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) : $sql;
231 my ($self, $options) = @_;
233 my $f = $options->{returning};
235 my $fieldlist = $self->_SWITCH_refkind($f, {
236 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
237 SCALAR => sub {$self->_quote($f)},
238 SCALARREF => sub {$$f},
240 return $self->_sqlcase(' returning ') . $fieldlist;
243 sub _insert_HASHREF { # explicit list of fields and then values
244 my ($self, $data) = @_;
246 my @fields = sort keys %$data;
248 my ($sql, @bind) = $self->_insert_values($data);
251 $_ = $self->_quote($_) foreach @fields;
252 $sql = "( ".join(", ", @fields).") ".$sql;
254 return ($sql, @bind);
257 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
258 my ($self, $data) = @_;
260 # no names (arrayref) so can't generate bindtype
261 $self->{bindtype} ne 'columns'
262 or belch "can't do 'columns' bindtype when called with arrayref";
264 # fold the list of values into a hash of column name - value pairs
265 # (where the column names are artificially generated, and their
266 # lexicographical ordering keep the ordering of the original list)
267 my $i = "a"; # incremented values will be in lexicographical order
268 my $data_in_hash = { map { ($i++ => $_) } @$data };
270 return $self->_insert_values($data_in_hash);
273 sub _insert_ARRAYREFREF { # literal SQL with bind
274 my ($self, $data) = @_;
276 my ($sql, @bind) = @${$data};
277 $self->_assert_bindval_matches_bindtype(@bind);
279 return ($sql, @bind);
283 sub _insert_SCALARREF { # literal SQL without bind
284 my ($self, $data) = @_;
290 my ($self, $data) = @_;
292 my (@values, @all_bind);
293 foreach my $column (sort keys %$data) {
294 my $v = $data->{$column};
296 $self->_SWITCH_refkind($v, {
299 if ($self->{array_datatypes}) { # if array datatype are activated
301 push @all_bind, $self->_bindtype($column, $v);
303 else { # else literal SQL with bind
304 my ($sql, @bind) = @$v;
305 $self->_assert_bindval_matches_bindtype(@bind);
307 push @all_bind, @bind;
311 ARRAYREFREF => sub { # literal SQL with bind
312 my ($sql, @bind) = @${$v};
313 $self->_assert_bindval_matches_bindtype(@bind);
315 push @all_bind, @bind;
318 # THINK : anything useful to do with a HASHREF ?
319 HASHREF => sub { # (nothing, but old SQLA passed it through)
320 #TODO in SQLA >= 2.0 it will die instead
321 belch "HASH ref as bind value in insert is not supported";
323 push @all_bind, $self->_bindtype($column, $v);
326 SCALARREF => sub { # literal SQL without bind
330 SCALAR_or_UNDEF => sub {
332 push @all_bind, $self->_bindtype($column, $v);
339 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
340 return ($sql, @all_bind);
345 #======================================================================
347 #======================================================================
352 my $table = $self->_table(shift);
353 my $data = shift || return;
357 # first build the 'SET' part of the sql statement
358 my (@set, @all_bind);
359 puke "Unsupported data type specified to \$sql->update"
360 unless ref $data eq 'HASH';
362 for my $k (sort keys %$data) {
365 my $label = $self->_quote($k);
367 $self->_SWITCH_refkind($v, {
369 if ($self->{array_datatypes}) { # array datatype
370 push @set, "$label = ?";
371 push @all_bind, $self->_bindtype($k, $v);
373 else { # literal SQL with bind
374 my ($sql, @bind) = @$v;
375 $self->_assert_bindval_matches_bindtype(@bind);
376 push @set, "$label = $sql";
377 push @all_bind, @bind;
380 ARRAYREFREF => sub { # literal SQL with bind
381 my ($sql, @bind) = @${$v};
382 $self->_assert_bindval_matches_bindtype(@bind);
383 push @set, "$label = $sql";
384 push @all_bind, @bind;
386 SCALARREF => sub { # literal SQL without bind
387 push @set, "$label = $$v";
390 my ($op, $arg, @rest) = %$v;
392 puke 'Operator calls in update must be in the form { -op => $arg }'
393 if (@rest or not $op =~ /^\-(.+)/);
395 local $self->{_nested_func_lhs} = $k;
396 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
398 push @set, "$label = $sql";
399 push @all_bind, @bind;
401 SCALAR_or_UNDEF => sub {
402 push @set, "$label = ?";
403 push @all_bind, $self->_bindtype($k, $v);
409 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
413 my($where_sql, @where_bind) = $self->where($where);
415 push @all_bind, @where_bind;
418 if ($options->{returning}) {
419 my ($returning_sql, @returning_bind) = $self->_returning ($options);
420 $sql .= $returning_sql;
421 push @all_bind, @returning_bind;
424 return wantarray ? ($sql, @all_bind) : $sql;
430 #======================================================================
432 #======================================================================
437 my $table = $self->_table(shift);
438 my $fields = shift || '*';
442 my($where_sql, @bind) = $self->where($where, $order);
444 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
446 my $sql = join(' ', $self->_sqlcase('select'), $f,
447 $self->_sqlcase('from'), $table)
450 return wantarray ? ($sql, @bind) : $sql;
453 #======================================================================
455 #======================================================================
460 my $table = $self->_table(shift);
464 my($where_sql, @bind) = $self->where($where);
465 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
467 return wantarray ? ($sql, @bind) : $sql;
471 #======================================================================
473 #======================================================================
477 # Finally, a separate routine just to handle WHERE clauses
479 my ($self, $where, $order) = @_;
482 my ($sql, @bind) = $self->_recurse_where($where);
483 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
487 $sql .= $self->_order_by($order);
490 return wantarray ? ($sql, @bind) : $sql;
495 my ($self, $where, $logic) = @_;
497 # dispatch on appropriate method according to refkind of $where
498 my $method = $self->_METHOD_FOR_refkind("_where", $where);
500 my ($sql, @bind) = $self->$method($where, $logic);
502 # DBIx::Class used to call _recurse_where in scalar context
503 # something else might too...
505 return ($sql, @bind);
508 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
515 #======================================================================
516 # WHERE: top-level ARRAYREF
517 #======================================================================
520 sub _where_ARRAYREF {
521 my ($self, $where, $logic) = @_;
523 $logic = uc($logic || $self->{logic});
524 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
526 my @clauses = @$where;
528 my (@sql_clauses, @all_bind);
529 # need to use while() so can shift() for pairs
531 my $el = shift @clauses;
533 $el = undef if (defined $el and ! length $el);
535 # switch according to kind of $el and get corresponding ($sql, @bind)
536 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
538 # skip empty elements, otherwise get invalid trailing AND stuff
539 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
543 $self->_assert_bindval_matches_bindtype(@b);
547 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
549 SCALARREF => sub { ($$el); },
552 # top-level arrayref with scalars, recurse in pairs
553 $self->_recurse_where({$el => shift(@clauses)})
556 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
560 push @sql_clauses, $sql;
561 push @all_bind, @bind;
565 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
568 #======================================================================
569 # WHERE: top-level ARRAYREFREF
570 #======================================================================
572 sub _where_ARRAYREFREF {
573 my ($self, $where) = @_;
574 my ($sql, @bind) = @$$where;
575 $self->_assert_bindval_matches_bindtype(@bind);
576 return ($sql, @bind);
579 #======================================================================
580 # WHERE: top-level HASHREF
581 #======================================================================
584 my ($self, $where) = @_;
585 my (@sql_clauses, @all_bind);
587 for my $k (sort keys %$where) {
588 my $v = $where->{$k};
590 # ($k => $v) is either a special unary op or a regular hashpair
591 my ($sql, @bind) = do {
593 # put the operator in canonical form
595 $op = substr $op, 1; # remove initial dash
596 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
597 $op =~ s/\s+/ /g; # compress whitespace
599 # so that -not_foo works correctly
600 $op =~ s/^not_/NOT /i;
602 $self->_debug("Unary OP(-$op) within hashref, recursing...");
603 my ($s, @b) = $self->_where_unary_op ($op, $v);
605 # top level vs nested
606 # we assume that handled unary ops will take care of their ()s
608 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
610 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
616 if (is_literal_value ($v) ) {
617 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
620 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
624 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
625 $self->$method($k, $v);
629 push @sql_clauses, $sql;
630 push @all_bind, @bind;
633 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
636 sub _where_unary_op {
637 my ($self, $op, $rhs) = @_;
639 # top level special ops are illegal in general
640 # this includes the -ident/-value ops (dual purpose unary and special)
641 puke "Illegal use of top-level '-$op'"
642 if ! defined $self->{_nested_func_lhs} and List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}};
644 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
645 my $handler = $op_entry->{handler};
647 if (not ref $handler) {
648 if ($op =~ s/ [_\s]? \d+ $//x ) {
649 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
650 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
652 return $self->$handler ($op, $rhs);
654 elsif (ref $handler eq 'CODE') {
655 return $handler->($self, $op, $rhs);
658 puke "Illegal handler for operator $op - expecting a method name or a coderef";
662 $self->_debug("Generic unary OP: $op - recursing as function");
664 $self->_assert_pass_injection_guard($op);
666 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
668 puke "Illegal use of top-level '-$op'"
669 unless defined $self->{_nested_func_lhs};
672 $self->_convert('?'),
673 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
677 $self->_recurse_where ($rhs)
681 $sql = sprintf ('%s %s',
682 $self->_sqlcase($op),
686 return ($sql, @bind);
689 sub _where_op_ANDOR {
690 my ($self, $op, $v) = @_;
692 $self->_SWITCH_refkind($v, {
694 return $self->_where_ARRAYREF($v, $op);
698 return ( $op =~ /^or/i )
699 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
700 : $self->_where_HASHREF($v);
704 puke "-$op => \\\$scalar makes little sense, use " .
706 ? '[ \$scalar, \%rest_of_conditions ] instead'
707 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
712 puke "-$op => \\[...] makes little sense, use " .
714 ? '[ \[...], \%rest_of_conditions ] instead'
715 : '-and => [ \[...], \%rest_of_conditions ] instead'
719 SCALAR => sub { # permissively interpreted as SQL
720 puke "-$op => \$value makes little sense, use -bool => \$value instead";
724 puke "-$op => undef not supported";
730 my ($self, $op, $v) = @_;
732 $self->_SWITCH_refkind($v, {
734 SCALAR => sub { # permissively interpreted as SQL
735 belch "literal SQL should be -nest => \\'scalar' "
736 . "instead of -nest => 'scalar' ";
741 puke "-$op => undef not supported";
745 $self->_recurse_where ($v);
753 my ($self, $op, $v) = @_;
755 my ($s, @b) = $self->_SWITCH_refkind($v, {
756 SCALAR => sub { # interpreted as SQL column
757 $self->_convert($self->_quote($v));
761 puke "-$op => undef not supported";
765 $self->_recurse_where ($v);
769 $s = "(NOT $s)" if $op =~ /^not/i;
774 sub _where_op_IDENT {
776 my ($op, $rhs) = splice @_, -2;
777 if (! defined $rhs or length ref $rhs) {
778 puke "-$op requires a single plain scalar argument (a quotable identifier)";
781 # in case we are called as a top level special op (no '=')
784 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
792 sub _where_op_VALUE {
794 my ($op, $rhs) = splice @_, -2;
796 # in case we are called as a top level special op (no '=')
800 if (! defined $rhs) {
802 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
809 ( defined $lhs ? $lhs : $self->{_nested_func_lhs} ),
816 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
820 $self->_convert('?'),
826 sub _where_hashpair_ARRAYREF {
827 my ($self, $k, $v) = @_;
830 my @v = @$v; # need copy because of shift below
831 $self->_debug("ARRAY($k) means distribute over elements");
833 # put apart first element if it is an operator (-and, -or)
835 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
839 my @distributed = map { {$k => $_} } @v;
842 $self->_debug("OP($op) reinjected into the distributed array");
843 unshift @distributed, $op;
846 my $logic = $op ? substr($op, 1) : '';
848 return $self->_recurse_where(\@distributed, $logic);
851 $self->_debug("empty ARRAY($k) means 0=1");
852 return ($self->{sqlfalse});
856 sub _where_hashpair_HASHREF {
857 my ($self, $k, $v, $logic) = @_;
860 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
861 ? $self->{_nested_func_lhs}
865 my ($all_sql, @all_bind);
867 for my $orig_op (sort keys %$v) {
868 my $val = $v->{$orig_op};
870 # put the operator in canonical form
873 # FIXME - we need to phase out dash-less ops
874 $op =~ s/^-//; # remove possible initial dash
875 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
876 $op =~ s/\s+/ /g; # compress whitespace
878 $self->_assert_pass_injection_guard($op);
881 $op =~ s/^is_not/IS NOT/i;
883 # so that -not_foo works correctly
884 $op =~ s/^not_/NOT /i;
886 # another retarded special case: foo => { $op => { -value => undef } }
887 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
893 # CASE: col-value logic modifiers
894 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
895 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
897 # CASE: special operators like -in or -between
898 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
899 my $handler = $special_op->{handler};
901 puke "No handler supplied for special operator $orig_op";
903 elsif (not ref $handler) {
904 ($sql, @bind) = $self->$handler ($k, $op, $val);
906 elsif (ref $handler eq 'CODE') {
907 ($sql, @bind) = $handler->($self, $k, $op, $val);
910 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
914 $self->_SWITCH_refkind($val, {
916 ARRAYREF => sub { # CASE: col => {op => \@vals}
917 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
920 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
921 my ($sub_sql, @sub_bind) = @$$val;
922 $self->_assert_bindval_matches_bindtype(@sub_bind);
923 $sql = join ' ', $self->_convert($self->_quote($k)),
924 $self->_sqlcase($op),
929 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
931 $op =~ /^not$/i ? 'is not' # legacy
932 : $op =~ $self->{equality_op} ? 'is'
933 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
934 : $op =~ $self->{inequality_op} ? 'is not'
935 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
936 : puke "unexpected operator '$orig_op' with undef operand";
938 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
941 FALLBACK => sub { # CASE: col => {op/func => $stuff}
942 ($sql, @bind) = $self->_where_unary_op ($op, $val);
945 $self->_convert($self->_quote($k)),
946 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
952 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
953 push @all_bind, @bind;
955 return ($all_sql, @all_bind);
958 sub _where_field_IS {
959 my ($self, $k, $op, $v) = @_;
961 my ($s) = $self->_SWITCH_refkind($v, {
964 $self->_convert($self->_quote($k)),
965 map { $self->_sqlcase($_)} ($op, 'null')
968 puke "$op can only take undef as argument";
975 sub _where_field_op_ARRAYREF {
976 my ($self, $k, $op, $vals) = @_;
978 my @vals = @$vals; #always work on a copy
981 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
983 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
986 # see if the first element is an -and/-or op
988 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
993 # a long standing API wart - an attempt to change this behavior during
994 # the 1.50 series failed *spectacularly*. Warn instead and leave the
999 (!$logic or $logic eq 'OR')
1001 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
1004 belch "A multi-element arrayref as an argument to the inequality op '$o' "
1005 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
1006 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
1010 # distribute $op over each remaining member of @vals, append logic if exists
1011 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1015 # try to DWIM on equality operators
1017 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1018 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1019 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1020 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1021 : puke "operator '$op' applied on an empty array (field '$k')";
1026 sub _where_hashpair_SCALARREF {
1027 my ($self, $k, $v) = @_;
1028 $self->_debug("SCALAR($k) means literal SQL: $$v");
1029 my $sql = $self->_quote($k) . " " . $$v;
1033 # literal SQL with bind
1034 sub _where_hashpair_ARRAYREFREF {
1035 my ($self, $k, $v) = @_;
1036 $self->_debug("REF($k) means literal SQL: @${$v}");
1037 my ($sql, @bind) = @$$v;
1038 $self->_assert_bindval_matches_bindtype(@bind);
1039 $sql = $self->_quote($k) . " " . $sql;
1040 return ($sql, @bind );
1043 # literal SQL without bind
1044 sub _where_hashpair_SCALAR {
1045 my ($self, $k, $v) = @_;
1046 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1047 my $sql = join ' ', $self->_convert($self->_quote($k)),
1048 $self->_sqlcase($self->{cmp}),
1049 $self->_convert('?');
1050 my @bind = $self->_bindtype($k, $v);
1051 return ( $sql, @bind);
1055 sub _where_hashpair_UNDEF {
1056 my ($self, $k, $v) = @_;
1057 $self->_debug("UNDEF($k) means IS NULL");
1058 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1062 #======================================================================
1063 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1064 #======================================================================
1067 sub _where_SCALARREF {
1068 my ($self, $where) = @_;
1071 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1077 my ($self, $where) = @_;
1080 $self->_debug("NOREF(*top) means literal SQL: $where");
1091 #======================================================================
1092 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1093 #======================================================================
1096 sub _where_field_BETWEEN {
1097 my ($self, $k, $op, $vals) = @_;
1099 my ($label, $and, $placeholder);
1100 $label = $self->_convert($self->_quote($k));
1101 $and = ' ' . $self->_sqlcase('and') . ' ';
1102 $placeholder = $self->_convert('?');
1103 $op = $self->_sqlcase($op);
1105 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1107 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1108 ARRAYREFREF => sub {
1109 my ($s, @b) = @$$vals;
1110 $self->_assert_bindval_matches_bindtype(@b);
1117 puke $invalid_args if @$vals != 2;
1119 my (@all_sql, @all_bind);
1120 foreach my $val (@$vals) {
1121 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1123 return ($placeholder, $self->_bindtype($k, $val) );
1128 ARRAYREFREF => sub {
1129 my ($sql, @bind) = @$$val;
1130 $self->_assert_bindval_matches_bindtype(@bind);
1131 return ($sql, @bind);
1134 my ($func, $arg, @rest) = %$val;
1135 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1136 if (@rest or $func !~ /^ \- (.+)/x);
1137 $self->_where_unary_op ($1 => $arg);
1143 push @all_sql, $sql;
1144 push @all_bind, @bind;
1148 (join $and, @all_sql),
1157 my $sql = "( $label $op $clause )";
1158 return ($sql, @bind)
1162 sub _where_field_IN {
1163 my ($self, $k, $op, $vals) = @_;
1165 # backwards compatibility : if scalar, force into an arrayref
1166 $vals = [$vals] if defined $vals && ! ref $vals;
1168 my ($label) = $self->_convert($self->_quote($k));
1169 my ($placeholder) = $self->_convert('?');
1170 $op = $self->_sqlcase($op);
1172 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1173 ARRAYREF => sub { # list of choices
1174 if (@$vals) { # nonempty list
1175 my (@all_sql, @all_bind);
1177 for my $val (@$vals) {
1178 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1180 return ($placeholder, $val);
1185 ARRAYREFREF => sub {
1186 my ($sql, @bind) = @$$val;
1187 $self->_assert_bindval_matches_bindtype(@bind);
1188 return ($sql, @bind);
1191 my ($func, $arg, @rest) = %$val;
1192 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1193 if (@rest or $func !~ /^ \- (.+)/x);
1194 $self->_where_unary_op ($1 => $arg);
1198 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1199 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1200 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1201 . 'will emit the logically correct SQL instead of raising this exception)'
1205 push @all_sql, $sql;
1206 push @all_bind, @bind;
1210 sprintf ('%s %s ( %s )',
1213 join (', ', @all_sql)
1215 $self->_bindtype($k, @all_bind),
1218 else { # empty list : some databases won't understand "IN ()", so DWIM
1219 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1224 SCALARREF => sub { # literal SQL
1225 my $sql = $self->_open_outer_paren ($$vals);
1226 return ("$label $op ( $sql )");
1228 ARRAYREFREF => sub { # literal SQL with bind
1229 my ($sql, @bind) = @$$vals;
1230 $self->_assert_bindval_matches_bindtype(@bind);
1231 $sql = $self->_open_outer_paren ($sql);
1232 return ("$label $op ( $sql )", @bind);
1236 puke "Argument passed to the '$op' operator can not be undefined";
1240 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1244 return ($sql, @bind);
1247 # Some databases (SQLite) treat col IN (1, 2) different from
1248 # col IN ( (1, 2) ). Use this to strip all outer parens while
1249 # adding them back in the corresponding method
1250 sub _open_outer_paren {
1251 my ($self, $sql) = @_;
1253 while ( my ($inner) = $sql =~ /^ \s* \( (.*) \) \s* $/xs ) {
1255 # there are closing parens inside, need the heavy duty machinery
1256 # to reevaluate the extraction starting from $sql (full reevaluation)
1257 if ( $inner =~ /\)/ ) {
1258 require Text::Balanced;
1260 my (undef, $remainder) = do {
1261 # idiotic design - writes to $@ but *DOES NOT* throw exceptions
1263 Text::Balanced::extract_bracketed( $sql, '()', qr/\s*/ );
1266 # the entire expression needs to be a balanced bracketed thing
1267 # (after an extract no remainder sans trailing space)
1268 last if defined $remainder and $remainder =~ /\S/;
1278 #======================================================================
1280 #======================================================================
1283 my ($self, $arg) = @_;
1286 for my $c ($self->_order_by_chunks ($arg) ) {
1287 $self->_SWITCH_refkind ($c, {
1288 SCALAR => sub { push @sql, $c },
1289 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1295 $self->_sqlcase(' order by'),
1301 return wantarray ? ($sql, @bind) : $sql;
1304 sub _order_by_chunks {
1305 my ($self, $arg) = @_;
1307 return $self->_SWITCH_refkind($arg, {
1310 map { $self->_order_by_chunks ($_ ) } @$arg;
1313 ARRAYREFREF => sub {
1314 my ($s, @b) = @$$arg;
1315 $self->_assert_bindval_matches_bindtype(@b);
1319 SCALAR => sub {$self->_quote($arg)},
1321 UNDEF => sub {return () },
1323 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1326 # get first pair in hash
1327 my ($key, $val, @rest) = %$arg;
1329 return () unless $key;
1331 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1332 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1338 for my $c ($self->_order_by_chunks ($val)) {
1341 $self->_SWITCH_refkind ($c, {
1346 ($sql, @bind) = @$c;
1350 $sql = $sql . ' ' . $self->_sqlcase($direction);
1352 push @ret, [ $sql, @bind];
1361 #======================================================================
1362 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1363 #======================================================================
1368 $self->_SWITCH_refkind($from, {
1369 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1370 SCALAR => sub {$self->_quote($from)},
1371 SCALARREF => sub {$$from},
1376 #======================================================================
1378 #======================================================================
1380 # highly optimized, as it's called way too often
1382 # my ($self, $label) = @_;
1384 return '' unless defined $_[1];
1385 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1387 unless ($_[0]->{quote_char}) {
1388 $_[0]->_assert_pass_injection_guard($_[1]);
1392 my $qref = ref $_[0]->{quote_char};
1395 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1397 elsif ($qref eq 'ARRAY') {
1398 ($l, $r) = @{$_[0]->{quote_char}};
1401 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1403 my $esc = $_[0]->{escape_char} || $r;
1405 # parts containing * are naturally unquoted
1406 return join( $_[0]->{name_sep}||'', map
1407 { $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } }
1408 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1413 # Conversion, if applicable
1415 #my ($self, $arg) = @_;
1416 if ($_[0]->{convert}) {
1417 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1424 #my ($self, $col, @vals) = @_;
1425 # called often - tighten code
1426 return $_[0]->{bindtype} eq 'columns'
1427 ? map {[$_[1], $_]} @_[2 .. $#_]
1432 # Dies if any element of @bind is not in [colname => value] format
1433 # if bindtype is 'columns'.
1434 sub _assert_bindval_matches_bindtype {
1435 # my ($self, @bind) = @_;
1437 if ($self->{bindtype} eq 'columns') {
1439 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1440 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1446 sub _join_sql_clauses {
1447 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1449 if (@$clauses_aref > 1) {
1450 my $join = " " . $self->_sqlcase($logic) . " ";
1451 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1452 return ($sql, @$bind_aref);
1454 elsif (@$clauses_aref) {
1455 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1458 return (); # if no SQL, ignore @$bind_aref
1463 # Fix SQL case, if so requested
1465 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1466 # don't touch the argument ... crooked logic, but let's not change it!
1467 return $_[0]->{case} ? $_[1] : uc($_[1]);
1471 #======================================================================
1472 # DISPATCHING FROM REFKIND
1473 #======================================================================
1476 my ($self, $data) = @_;
1478 return 'UNDEF' unless defined $data;
1480 # blessed objects are treated like scalars
1481 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1483 return 'SCALAR' unless $ref;
1486 while ($ref eq 'REF') {
1488 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1492 return ($ref||'SCALAR') . ('REF' x $n_steps);
1496 my ($self, $data) = @_;
1497 my @try = ($self->_refkind($data));
1498 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1499 push @try, 'FALLBACK';
1503 sub _METHOD_FOR_refkind {
1504 my ($self, $meth_prefix, $data) = @_;
1507 for (@{$self->_try_refkind($data)}) {
1508 $method = $self->can($meth_prefix."_".$_)
1512 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1516 sub _SWITCH_refkind {
1517 my ($self, $data, $dispatch_table) = @_;
1520 for (@{$self->_try_refkind($data)}) {
1521 $coderef = $dispatch_table->{$_}
1525 puke "no dispatch entry for ".$self->_refkind($data)
1534 #======================================================================
1535 # VALUES, GENERATE, AUTOLOAD
1536 #======================================================================
1538 # LDNOTE: original code from nwiger, didn't touch code in that section
1539 # I feel the AUTOLOAD stuff should not be the default, it should
1540 # only be activated on explicit demand by user.
1544 my $data = shift || return;
1545 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1546 unless ref $data eq 'HASH';
1549 foreach my $k ( sort keys %$data ) {
1550 my $v = $data->{$k};
1551 $self->_SWITCH_refkind($v, {
1553 if ($self->{array_datatypes}) { # array datatype
1554 push @all_bind, $self->_bindtype($k, $v);
1556 else { # literal SQL with bind
1557 my ($sql, @bind) = @$v;
1558 $self->_assert_bindval_matches_bindtype(@bind);
1559 push @all_bind, @bind;
1562 ARRAYREFREF => sub { # literal SQL with bind
1563 my ($sql, @bind) = @${$v};
1564 $self->_assert_bindval_matches_bindtype(@bind);
1565 push @all_bind, @bind;
1567 SCALARREF => sub { # literal SQL without bind
1569 SCALAR_or_UNDEF => sub {
1570 push @all_bind, $self->_bindtype($k, $v);
1581 my(@sql, @sqlq, @sqlv);
1585 if ($ref eq 'HASH') {
1586 for my $k (sort keys %$_) {
1589 my $label = $self->_quote($k);
1590 if ($r eq 'ARRAY') {
1591 # literal SQL with bind
1592 my ($sql, @bind) = @$v;
1593 $self->_assert_bindval_matches_bindtype(@bind);
1594 push @sqlq, "$label = $sql";
1596 } elsif ($r eq 'SCALAR') {
1597 # literal SQL without bind
1598 push @sqlq, "$label = $$v";
1600 push @sqlq, "$label = ?";
1601 push @sqlv, $self->_bindtype($k, $v);
1604 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1605 } elsif ($ref eq 'ARRAY') {
1606 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1609 if ($r eq 'ARRAY') { # literal SQL with bind
1610 my ($sql, @bind) = @$v;
1611 $self->_assert_bindval_matches_bindtype(@bind);
1614 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1615 # embedded literal SQL
1622 push @sql, '(' . join(', ', @sqlq) . ')';
1623 } elsif ($ref eq 'SCALAR') {
1627 # strings get case twiddled
1628 push @sql, $self->_sqlcase($_);
1632 my $sql = join ' ', @sql;
1634 # this is pretty tricky
1635 # if ask for an array, return ($stmt, @bind)
1636 # otherwise, s/?/shift @sqlv/ to put it inline
1638 return ($sql, @sqlv);
1640 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1641 ref $d ? $d->[1] : $d/e;
1650 # This allows us to check for a local, then _form, attr
1652 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1653 return $self->generate($name, @_);
1664 SQL::Abstract - Generate SQL from Perl data structures
1670 my $sql = SQL::Abstract->new;
1672 my($stmt, @bind) = $sql->select($source, \@fields, \%where, \@order);
1674 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1676 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1678 my($stmt, @bind) = $sql->delete($table, \%where);
1680 # Then, use these in your DBI statements
1681 my $sth = $dbh->prepare($stmt);
1682 $sth->execute(@bind);
1684 # Just generate the WHERE clause
1685 my($stmt, @bind) = $sql->where(\%where, \@order);
1687 # Return values in the same order, for hashed queries
1688 # See PERFORMANCE section for more details
1689 my @bind = $sql->values(\%fieldvals);
1693 This module was inspired by the excellent L<DBIx::Abstract>.
1694 However, in using that module I found that what I really wanted
1695 to do was generate SQL, but still retain complete control over my
1696 statement handles and use the DBI interface. So, I set out to
1697 create an abstract SQL generation module.
1699 While based on the concepts used by L<DBIx::Abstract>, there are
1700 several important differences, especially when it comes to WHERE
1701 clauses. I have modified the concepts used to make the SQL easier
1702 to generate from Perl data structures and, IMO, more intuitive.
1703 The underlying idea is for this module to do what you mean, based
1704 on the data structures you provide it. The big advantage is that
1705 you don't have to modify your code every time your data changes,
1706 as this module figures it out.
1708 To begin with, an SQL INSERT is as easy as just specifying a hash
1709 of C<key=value> pairs:
1712 name => 'Jimbo Bobson',
1713 phone => '123-456-7890',
1714 address => '42 Sister Lane',
1715 city => 'St. Louis',
1716 state => 'Louisiana',
1719 The SQL can then be generated with this:
1721 my($stmt, @bind) = $sql->insert('people', \%data);
1723 Which would give you something like this:
1725 $stmt = "INSERT INTO people
1726 (address, city, name, phone, state)
1727 VALUES (?, ?, ?, ?, ?)";
1728 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1729 '123-456-7890', 'Louisiana');
1731 These are then used directly in your DBI code:
1733 my $sth = $dbh->prepare($stmt);
1734 $sth->execute(@bind);
1736 =head2 Inserting and Updating Arrays
1738 If your database has array types (like for example Postgres),
1739 activate the special option C<< array_datatypes => 1 >>
1740 when creating the C<SQL::Abstract> object.
1741 Then you may use an arrayref to insert and update database array types:
1743 my $sql = SQL::Abstract->new(array_datatypes => 1);
1745 planets => [qw/Mercury Venus Earth Mars/]
1748 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1752 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1754 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1757 =head2 Inserting and Updating SQL
1759 In order to apply SQL functions to elements of your C<%data> you may
1760 specify a reference to an arrayref for the given hash value. For example,
1761 if you need to execute the Oracle C<to_date> function on a value, you can
1762 say something like this:
1766 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1769 The first value in the array is the actual SQL. Any other values are
1770 optional and would be included in the bind values array. This gives
1773 my($stmt, @bind) = $sql->insert('people', \%data);
1775 $stmt = "INSERT INTO people (name, date_entered)
1776 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1777 @bind = ('Bill', '03/02/2003');
1779 An UPDATE is just as easy, all you change is the name of the function:
1781 my($stmt, @bind) = $sql->update('people', \%data);
1783 Notice that your C<%data> isn't touched; the module will generate
1784 the appropriately quirky SQL for you automatically. Usually you'll
1785 want to specify a WHERE clause for your UPDATE, though, which is
1786 where handling C<%where> hashes comes in handy...
1788 =head2 Complex where statements
1790 This module can generate pretty complicated WHERE statements
1791 easily. For example, simple C<key=value> pairs are taken to mean
1792 equality, and if you want to see if a field is within a set
1793 of values, you can use an arrayref. Let's say we wanted to
1794 SELECT some data based on this criteria:
1797 requestor => 'inna',
1798 worker => ['nwiger', 'rcwe', 'sfz'],
1799 status => { '!=', 'completed' }
1802 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1804 The above would give you something like this:
1806 $stmt = "SELECT * FROM tickets WHERE
1807 ( requestor = ? ) AND ( status != ? )
1808 AND ( worker = ? OR worker = ? OR worker = ? )";
1809 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1811 Which you could then use in DBI code like so:
1813 my $sth = $dbh->prepare($stmt);
1814 $sth->execute(@bind);
1820 The methods are simple. There's one for every major SQL operation,
1821 and a constructor you use first. The arguments are specified in a
1822 similar order for each method (table, then fields, then a where
1823 clause) to try and simplify things.
1825 =head2 new(option => 'value')
1827 The C<new()> function takes a list of options and values, and returns
1828 a new B<SQL::Abstract> object which can then be used to generate SQL
1829 through the methods below. The options accepted are:
1835 If set to 'lower', then SQL will be generated in all lowercase. By
1836 default SQL is generated in "textbook" case meaning something like:
1838 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1840 Any setting other than 'lower' is ignored.
1844 This determines what the default comparison operator is. By default
1845 it is C<=>, meaning that a hash like this:
1847 %where = (name => 'nwiger', email => 'nate@wiger.org');
1849 Will generate SQL like this:
1851 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1853 However, you may want loose comparisons by default, so if you set
1854 C<cmp> to C<like> you would get SQL such as:
1856 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1858 You can also override the comparison on an individual basis - see
1859 the huge section on L</"WHERE CLAUSES"> at the bottom.
1861 =item sqltrue, sqlfalse
1863 Expressions for inserting boolean values within SQL statements.
1864 By default these are C<1=1> and C<1=0>. They are used
1865 by the special operators C<-in> and C<-not_in> for generating
1866 correct SQL even when the argument is an empty array (see below).
1870 This determines the default logical operator for multiple WHERE
1871 statements in arrays or hashes. If absent, the default logic is "or"
1872 for arrays, and "and" for hashes. This means that a WHERE
1876 event_date => {'>=', '2/13/99'},
1877 event_date => {'<=', '4/24/03'},
1880 will generate SQL like this:
1882 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1884 This is probably not what you want given this query, though (look
1885 at the dates). To change the "OR" to an "AND", simply specify:
1887 my $sql = SQL::Abstract->new(logic => 'and');
1889 Which will change the above C<WHERE> to:
1891 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1893 The logic can also be changed locally by inserting
1894 a modifier in front of an arrayref :
1896 @where = (-and => [event_date => {'>=', '2/13/99'},
1897 event_date => {'<=', '4/24/03'} ]);
1899 See the L</"WHERE CLAUSES"> section for explanations.
1903 This will automatically convert comparisons using the specified SQL
1904 function for both column and value. This is mostly used with an argument
1905 of C<upper> or C<lower>, so that the SQL will have the effect of
1906 case-insensitive "searches". For example, this:
1908 $sql = SQL::Abstract->new(convert => 'upper');
1909 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1911 Will turn out the following SQL:
1913 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1915 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1916 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1917 not validate this option; it will just pass through what you specify verbatim).
1921 This is a kludge because many databases suck. For example, you can't
1922 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1923 Instead, you have to use C<bind_param()>:
1925 $sth->bind_param(1, 'reg data');
1926 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1928 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1929 which loses track of which field each slot refers to. Fear not.
1931 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1932 Currently, you can specify either C<normal> (default) or C<columns>. If you
1933 specify C<columns>, you will get an array that looks like this:
1935 my $sql = SQL::Abstract->new(bindtype => 'columns');
1936 my($stmt, @bind) = $sql->insert(...);
1939 [ 'column1', 'value1' ],
1940 [ 'column2', 'value2' ],
1941 [ 'column3', 'value3' ],
1944 You can then iterate through this manually, using DBI's C<bind_param()>.
1946 $sth->prepare($stmt);
1949 my($col, $data) = @$_;
1950 if ($col eq 'details' || $col eq 'comments') {
1951 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1952 } elsif ($col eq 'image') {
1953 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1955 $sth->bind_param($i, $data);
1959 $sth->execute; # execute without @bind now
1961 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1962 Basically, the advantage is still that you don't have to care which fields
1963 are or are not included. You could wrap that above C<for> loop in a simple
1964 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1965 get a layer of abstraction over manual SQL specification.
1967 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1968 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1969 will expect the bind values in this format.
1973 This is the character that a table or column name will be quoted
1974 with. By default this is an empty string, but you could set it to
1975 the character C<`>, to generate SQL like this:
1977 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1979 Alternatively, you can supply an array ref of two items, the first being the left
1980 hand quote character, and the second the right hand quote character. For
1981 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1982 that generates SQL like this:
1984 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1986 Quoting is useful if you have tables or columns names that are reserved
1987 words in your database's SQL dialect.
1991 This is the character that will be used to escape L</quote_char>s appearing
1992 in an identifier before it has been quoted.
1994 The parameter default in case of a single L</quote_char> character is the quote
1997 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1998 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
1999 of the B<opening (left)> L</quote_char> within the identifier are currently left
2000 untouched. The default for opening-closing-style quotes may change in future
2001 versions, thus you are B<strongly encouraged> to specify the escape character
2006 This is the character that separates a table and column name. It is
2007 necessary to specify this when the C<quote_char> option is selected,
2008 so that tables and column names can be individually quoted like this:
2010 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
2012 =item injection_guard
2014 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
2015 column name specified in a query structure. This is a safety mechanism to avoid
2016 injection attacks when mishandling user input e.g.:
2018 my %condition_as_column_value_pairs = get_values_from_user();
2019 $sqla->select( ... , \%condition_as_column_value_pairs );
2021 If the expression matches an exception is thrown. Note that literal SQL
2022 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
2024 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
2026 =item array_datatypes
2028 When this option is true, arrayrefs in INSERT or UPDATE are
2029 interpreted as array datatypes and are passed directly
2031 When this option is false, arrayrefs are interpreted
2032 as literal SQL, just like refs to arrayrefs
2033 (but this behavior is for backwards compatibility; when writing
2034 new queries, use the "reference to arrayref" syntax
2040 Takes a reference to a list of "special operators"
2041 to extend the syntax understood by L<SQL::Abstract>.
2042 See section L</"SPECIAL OPERATORS"> for details.
2046 Takes a reference to a list of "unary operators"
2047 to extend the syntax understood by L<SQL::Abstract>.
2048 See section L</"UNARY OPERATORS"> for details.
2054 =head2 insert($table, \@values || \%fieldvals, \%options)
2056 This is the simplest function. You simply give it a table name
2057 and either an arrayref of values or hashref of field/value pairs.
2058 It returns an SQL INSERT statement and a list of bind values.
2059 See the sections on L</"Inserting and Updating Arrays"> and
2060 L</"Inserting and Updating SQL"> for information on how to insert
2061 with those data types.
2063 The optional C<\%options> hash reference may contain additional
2064 options to generate the insert SQL. Currently supported options
2071 Takes either a scalar of raw SQL fields, or an array reference of
2072 field names, and adds on an SQL C<RETURNING> statement at the end.
2073 This allows you to return data generated by the insert statement
2074 (such as row IDs) without performing another C<SELECT> statement.
2075 Note, however, this is not part of the SQL standard and may not
2076 be supported by all database engines.
2080 =head2 update($table, \%fieldvals, \%where, \%options)
2082 This takes a table, hashref of field/value pairs, and an optional
2083 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2085 See the sections on L</"Inserting and Updating Arrays"> and
2086 L</"Inserting and Updating SQL"> for information on how to insert
2087 with those data types.
2089 The optional C<\%options> hash reference may contain additional
2090 options to generate the update SQL. Currently supported options
2097 See the C<returning> option to
2098 L<insert|/insert($table, \@values || \%fieldvals, \%options)>.
2102 =head2 select($source, $fields, $where, $order)
2104 This returns a SQL SELECT statement and associated list of bind values, as
2105 specified by the arguments :
2111 Specification of the 'FROM' part of the statement.
2112 The argument can be either a plain scalar (interpreted as a table
2113 name, will be quoted), or an arrayref (interpreted as a list
2114 of table names, joined by commas, quoted), or a scalarref
2115 (literal table name, not quoted), or a ref to an arrayref
2116 (list of literal table names, joined by commas, not quoted).
2120 Specification of the list of fields to retrieve from
2122 The argument can be either an arrayref (interpreted as a list
2123 of field names, will be joined by commas and quoted), or a
2124 plain scalar (literal SQL, not quoted).
2125 Please observe that this API is not as flexible as that of
2126 the first argument C<$source>, for backwards compatibility reasons.
2130 Optional argument to specify the WHERE part of the query.
2131 The argument is most often a hashref, but can also be
2132 an arrayref or plain scalar --
2133 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2137 Optional argument to specify the ORDER BY part of the query.
2138 The argument can be a scalar, a hashref or an arrayref
2139 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2145 =head2 delete($table, \%where)
2147 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2148 It returns an SQL DELETE statement and list of bind values.
2150 =head2 where(\%where, \@order)
2152 This is used to generate just the WHERE clause. For example,
2153 if you have an arbitrary data structure and know what the
2154 rest of your SQL is going to look like, but want an easy way
2155 to produce a WHERE clause, use this. It returns an SQL WHERE
2156 clause and list of bind values.
2159 =head2 values(\%data)
2161 This just returns the values from the hash C<%data>, in the same
2162 order that would be returned from any of the other above queries.
2163 Using this allows you to markedly speed up your queries if you
2164 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2166 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2168 Warning: This is an experimental method and subject to change.
2170 This returns arbitrarily generated SQL. It's a really basic shortcut.
2171 It will return two different things, depending on return context:
2173 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2174 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2176 These would return the following:
2178 # First calling form
2179 $stmt = "CREATE TABLE test (?, ?)";
2180 @bind = (field1, field2);
2182 # Second calling form
2183 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2185 Depending on what you're trying to do, it's up to you to choose the correct
2186 format. In this example, the second form is what you would want.
2190 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2194 ALTER SESSION SET nls_date_format = 'MM/YY'
2196 You get the idea. Strings get their case twiddled, but everything
2197 else remains verbatim.
2199 =head1 EXPORTABLE FUNCTIONS
2201 =head2 is_plain_value
2203 Determines if the supplied argument is a plain value as understood by this
2208 =item * The value is C<undef>
2210 =item * The value is a non-reference
2212 =item * The value is an object with stringification overloading
2214 =item * The value is of the form C<< { -value => $anything } >>
2218 On failure returns C<undef>, on sucess returns a B<scalar> reference
2219 to the original supplied argument.
2225 The stringification overloading detection is rather advanced: it takes
2226 into consideration not only the presence of a C<""> overload, but if that
2227 fails also checks for enabled
2228 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2229 on either C<0+> or C<bool>.
2231 Unfortunately testing in the field indicates that this
2232 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2233 but only when very large numbers of stringifying objects are involved.
2234 At the time of writing ( Sep 2014 ) there is no clear explanation of
2235 the direct cause, nor is there a manageably small test case that reliably
2236 reproduces the problem.
2238 If you encounter any of the following exceptions in B<random places within
2239 your application stack> - this module may be to blame:
2241 Operation "ne": no method found,
2242 left argument in overloaded package <something>,
2243 right argument in overloaded package <something>
2247 Stub found while resolving method "???" overloading """" in package <something>
2249 If you fall victim to the above - please attempt to reduce the problem
2250 to something that could be sent to the L<SQL::Abstract developers
2251 |DBIx::Class/GETTING HELP/SUPPORT>
2252 (either publicly or privately). As a workaround in the meantime you can
2253 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2254 value, which will most likely eliminate your problem (at the expense of
2255 not being able to properly detect exotic forms of stringification).
2257 This notice and environment variable will be removed in a future version,
2258 as soon as the underlying problem is found and a reliable workaround is
2263 =head2 is_literal_value
2265 Determines if the supplied argument is a literal value as understood by this
2270 =item * C<\$sql_string>
2272 =item * C<\[ $sql_string, @bind_values ]>
2276 On failure returns C<undef>, on sucess returns an B<array> reference
2277 containing the unpacked version of the supplied literal SQL and bind values.
2279 =head1 WHERE CLAUSES
2283 This module uses a variation on the idea from L<DBIx::Abstract>. It
2284 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2285 module is that things in arrays are OR'ed, and things in hashes
2288 The easiest way to explain is to show lots of examples. After
2289 each C<%where> hash shown, it is assumed you used:
2291 my($stmt, @bind) = $sql->where(\%where);
2293 However, note that the C<%where> hash can be used directly in any
2294 of the other functions as well, as described above.
2296 =head2 Key-value pairs
2298 So, let's get started. To begin, a simple hash:
2302 status => 'completed'
2305 Is converted to SQL C<key = val> statements:
2307 $stmt = "WHERE user = ? AND status = ?";
2308 @bind = ('nwiger', 'completed');
2310 One common thing I end up doing is having a list of values that
2311 a field can be in. To do this, simply specify a list inside of
2316 status => ['assigned', 'in-progress', 'pending'];
2319 This simple code will create the following:
2321 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2322 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2324 A field associated to an empty arrayref will be considered a
2325 logical false and will generate 0=1.
2327 =head2 Tests for NULL values
2329 If the value part is C<undef> then this is converted to SQL <IS NULL>
2338 $stmt = "WHERE user = ? AND status IS NULL";
2341 To test if a column IS NOT NULL:
2345 status => { '!=', undef },
2348 =head2 Specific comparison operators
2350 If you want to specify a different type of operator for your comparison,
2351 you can use a hashref for a given column:
2355 status => { '!=', 'completed' }
2358 Which would generate:
2360 $stmt = "WHERE user = ? AND status != ?";
2361 @bind = ('nwiger', 'completed');
2363 To test against multiple values, just enclose the values in an arrayref:
2365 status => { '=', ['assigned', 'in-progress', 'pending'] };
2367 Which would give you:
2369 "WHERE status = ? OR status = ? OR status = ?"
2372 The hashref can also contain multiple pairs, in which case it is expanded
2373 into an C<AND> of its elements:
2377 status => { '!=', 'completed', -not_like => 'pending%' }
2380 # Or more dynamically, like from a form
2381 $where{user} = 'nwiger';
2382 $where{status}{'!='} = 'completed';
2383 $where{status}{'-not_like'} = 'pending%';
2385 # Both generate this
2386 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2387 @bind = ('nwiger', 'completed', 'pending%');
2390 To get an OR instead, you can combine it with the arrayref idea:
2394 priority => [ { '=', 2 }, { '>', 5 } ]
2397 Which would generate:
2399 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2400 @bind = ('2', '5', 'nwiger');
2402 If you want to include literal SQL (with or without bind values), just use a
2403 scalar reference or reference to an arrayref as the value:
2406 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2407 date_expires => { '<' => \"now()" }
2410 Which would generate:
2412 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2413 @bind = ('11/26/2008');
2416 =head2 Logic and nesting operators
2418 In the example above,
2419 there is a subtle trap if you want to say something like
2420 this (notice the C<AND>):
2422 WHERE priority != ? AND priority != ?
2424 Because, in Perl you I<can't> do this:
2426 priority => { '!=' => 2, '!=' => 1 }
2428 As the second C<!=> key will obliterate the first. The solution
2429 is to use the special C<-modifier> form inside an arrayref:
2431 priority => [ -and => {'!=', 2},
2435 Normally, these would be joined by C<OR>, but the modifier tells it
2436 to use C<AND> instead. (Hint: You can use this in conjunction with the
2437 C<logic> option to C<new()> in order to change the way your queries
2438 work by default.) B<Important:> Note that the C<-modifier> goes
2439 B<INSIDE> the arrayref, as an extra first element. This will
2440 B<NOT> do what you think it might:
2442 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2444 Here is a quick list of equivalencies, since there is some overlap:
2447 status => {'!=', 'completed', 'not like', 'pending%' }
2448 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2451 status => {'=', ['assigned', 'in-progress']}
2452 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2453 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2457 =head2 Special operators : IN, BETWEEN, etc.
2459 You can also use the hashref format to compare a list of fields using the
2460 C<IN> comparison operator, by specifying the list as an arrayref:
2463 status => 'completed',
2464 reportid => { -in => [567, 2335, 2] }
2467 Which would generate:
2469 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2470 @bind = ('completed', '567', '2335', '2');
2472 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2475 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2476 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2477 'sqltrue' (by default : C<1=1>).
2479 In addition to the array you can supply a chunk of literal sql or
2480 literal sql with bind:
2483 customer => { -in => \[
2484 'SELECT cust_id FROM cust WHERE balance > ?',
2487 status => { -in => \'SELECT status_codes FROM states' },
2493 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2494 AND status IN ( SELECT status_codes FROM states )
2498 Finally, if the argument to C<-in> is not a reference, it will be
2499 treated as a single-element array.
2501 Another pair of operators is C<-between> and C<-not_between>,
2502 used with an arrayref of two values:
2506 completion_date => {
2507 -not_between => ['2002-10-01', '2003-02-06']
2513 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2515 Just like with C<-in> all plausible combinations of literal SQL
2519 start0 => { -between => [ 1, 2 ] },
2520 start1 => { -between => \["? AND ?", 1, 2] },
2521 start2 => { -between => \"lower(x) AND upper(y)" },
2522 start3 => { -between => [
2524 \["upper(?)", 'stuff' ],
2531 ( start0 BETWEEN ? AND ? )
2532 AND ( start1 BETWEEN ? AND ? )
2533 AND ( start2 BETWEEN lower(x) AND upper(y) )
2534 AND ( start3 BETWEEN lower(x) AND upper(?) )
2536 @bind = (1, 2, 1, 2, 'stuff');
2539 These are the two builtin "special operators"; but the
2540 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2542 =head2 Unary operators: bool
2544 If you wish to test against boolean columns or functions within your
2545 database you can use the C<-bool> and C<-not_bool> operators. For
2546 example to test the column C<is_user> being true and the column
2547 C<is_enabled> being false you would use:-
2551 -not_bool => 'is_enabled',
2556 WHERE is_user AND NOT is_enabled
2558 If a more complex combination is required, testing more conditions,
2559 then you should use the and/or operators:-
2564 -not_bool => { two=> { -rlike => 'bar' } },
2565 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2576 (NOT ( three = ? OR three > ? ))
2579 =head2 Nested conditions, -and/-or prefixes
2581 So far, we've seen how multiple conditions are joined with a top-level
2582 C<AND>. We can change this by putting the different conditions we want in
2583 hashes and then putting those hashes in an array. For example:
2588 status => { -like => ['pending%', 'dispatched'] },
2592 status => 'unassigned',
2596 This data structure would create the following:
2598 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2599 OR ( user = ? AND status = ? ) )";
2600 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2603 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2604 to change the logic inside :
2610 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2611 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2618 $stmt = "WHERE ( user = ?
2619 AND ( ( workhrs > ? AND geo = ? )
2620 OR ( workhrs < ? OR geo = ? ) ) )";
2621 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2623 =head3 Algebraic inconsistency, for historical reasons
2625 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2626 operator goes C<outside> of the nested structure; whereas when connecting
2627 several constraints on one column, the C<-and> operator goes
2628 C<inside> the arrayref. Here is an example combining both features :
2631 -and => [a => 1, b => 2],
2632 -or => [c => 3, d => 4],
2633 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2638 WHERE ( ( ( a = ? AND b = ? )
2639 OR ( c = ? OR d = ? )
2640 OR ( e LIKE ? AND e LIKE ? ) ) )
2642 This difference in syntax is unfortunate but must be preserved for
2643 historical reasons. So be careful : the two examples below would
2644 seem algebraically equivalent, but they are not
2646 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2647 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2649 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2650 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2653 =head2 Literal SQL and value type operators
2655 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2656 side" is a column name and the "right side" is a value (normally rendered as
2657 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2658 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2659 alter this behavior. There are several ways of doing so.
2663 This is a virtual operator that signals the string to its right side is an
2664 identifier (a column name) and not a value. For example to compare two
2665 columns you would write:
2668 priority => { '<', 2 },
2669 requestor => { -ident => 'submitter' },
2674 $stmt = "WHERE priority < ? AND requestor = submitter";
2677 If you are maintaining legacy code you may see a different construct as
2678 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2683 This is a virtual operator that signals that the construct to its right side
2684 is a value to be passed to DBI. This is for example necessary when you want
2685 to write a where clause against an array (for RDBMS that support such
2686 datatypes). For example:
2689 array => { -value => [1, 2, 3] }
2694 $stmt = 'WHERE array = ?';
2695 @bind = ([1, 2, 3]);
2697 Note that if you were to simply say:
2703 the result would probably not be what you wanted:
2705 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2710 Finally, sometimes only literal SQL will do. To include a random snippet
2711 of SQL verbatim, you specify it as a scalar reference. Consider this only
2712 as a last resort. Usually there is a better way. For example:
2715 priority => { '<', 2 },
2716 requestor => { -in => \'(SELECT name FROM hitmen)' },
2721 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2724 Note that in this example, you only get one bind parameter back, since
2725 the verbatim SQL is passed as part of the statement.
2729 Never use untrusted input as a literal SQL argument - this is a massive
2730 security risk (there is no way to check literal snippets for SQL
2731 injections and other nastyness). If you need to deal with untrusted input
2732 use literal SQL with placeholders as described next.
2734 =head3 Literal SQL with placeholders and bind values (subqueries)
2736 If the literal SQL to be inserted has placeholders and bind values,
2737 use a reference to an arrayref (yes this is a double reference --
2738 not so common, but perfectly legal Perl). For example, to find a date
2739 in Postgres you can use something like this:
2742 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2747 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2750 Note that you must pass the bind values in the same format as they are returned
2751 by L<where|/where(\%where, \@order)>. This means that if you set L</bindtype>
2752 to C<columns>, you must provide the bind values in the
2753 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2754 scalar value; most commonly the column name, but you can use any scalar value
2755 (including references and blessed references), L<SQL::Abstract> will simply
2756 pass it through intact. So if C<bindtype> is set to C<columns> the above
2757 example will look like:
2760 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2763 Literal SQL is especially useful for nesting parenthesized clauses in the
2764 main SQL query. Here is a first example :
2766 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2770 bar => \["IN ($sub_stmt)" => @sub_bind],
2775 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2776 WHERE c2 < ? AND c3 LIKE ?))";
2777 @bind = (1234, 100, "foo%");
2779 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2780 are expressed in the same way. Of course the C<$sub_stmt> and
2781 its associated bind values can be generated through a former call
2784 my ($sub_stmt, @sub_bind)
2785 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2786 c3 => {-like => "foo%"}});
2789 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2792 In the examples above, the subquery was used as an operator on a column;
2793 but the same principle also applies for a clause within the main C<%where>
2794 hash, like an EXISTS subquery :
2796 my ($sub_stmt, @sub_bind)
2797 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2798 my %where = ( -and => [
2800 \["EXISTS ($sub_stmt)" => @sub_bind],
2805 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2806 WHERE c1 = ? AND c2 > t0.c0))";
2810 Observe that the condition on C<c2> in the subquery refers to
2811 column C<t0.c0> of the main query : this is I<not> a bind
2812 value, so we have to express it through a scalar ref.
2813 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2814 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2815 what we wanted here.
2817 Finally, here is an example where a subquery is used
2818 for expressing unary negation:
2820 my ($sub_stmt, @sub_bind)
2821 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2822 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2824 lname => {like => '%son%'},
2825 \["NOT ($sub_stmt)" => @sub_bind],
2830 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2831 @bind = ('%son%', 10, 20)
2833 =head3 Deprecated usage of Literal SQL
2835 Below are some examples of archaic use of literal SQL. It is shown only as
2836 reference for those who deal with legacy code. Each example has a much
2837 better, cleaner and safer alternative that users should opt for in new code.
2843 my %where = ( requestor => \'IS NOT NULL' )
2845 $stmt = "WHERE requestor IS NOT NULL"
2847 This used to be the way of generating NULL comparisons, before the handling
2848 of C<undef> got formalized. For new code please use the superior syntax as
2849 described in L</Tests for NULL values>.
2853 my %where = ( requestor => \'= submitter' )
2855 $stmt = "WHERE requestor = submitter"
2857 This used to be the only way to compare columns. Use the superior L</-ident>
2858 method for all new code. For example an identifier declared in such a way
2859 will be properly quoted if L</quote_char> is properly set, while the legacy
2860 form will remain as supplied.
2864 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2866 $stmt = "WHERE completed > ? AND is_ready"
2867 @bind = ('2012-12-21')
2869 Using an empty string literal used to be the only way to express a boolean.
2870 For all new code please use the much more readable
2871 L<-bool|/Unary operators: bool> operator.
2877 These pages could go on for a while, since the nesting of the data
2878 structures this module can handle are pretty much unlimited (the
2879 module implements the C<WHERE> expansion as a recursive function
2880 internally). Your best bet is to "play around" with the module a
2881 little to see how the data structures behave, and choose the best
2882 format for your data based on that.
2884 And of course, all the values above will probably be replaced with
2885 variables gotten from forms or the command line. After all, if you
2886 knew everything ahead of time, you wouldn't have to worry about
2887 dynamically-generating SQL and could just hardwire it into your
2890 =head1 ORDER BY CLAUSES
2892 Some functions take an order by clause. This can either be a scalar (just a
2893 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2894 or an array of either of the two previous forms. Examples:
2896 Given | Will Generate
2897 ----------------------------------------------------------
2899 \'colA DESC' | ORDER BY colA DESC
2901 'colA' | ORDER BY colA
2903 [qw/colA colB/] | ORDER BY colA, colB
2905 {-asc => 'colA'} | ORDER BY colA ASC
2907 {-desc => 'colB'} | ORDER BY colB DESC
2909 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2911 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2914 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2915 { -desc => [qw/colB/], | colC ASC, colD ASC
2916 { -asc => [qw/colC colD/],|
2918 ===========================================================
2922 =head1 SPECIAL OPERATORS
2924 my $sqlmaker = SQL::Abstract->new(special_ops => [
2928 my ($self, $field, $op, $arg) = @_;
2934 handler => 'method_name',
2938 A "special operator" is a SQL syntactic clause that can be
2939 applied to a field, instead of a usual binary operator.
2942 WHERE field IN (?, ?, ?)
2943 WHERE field BETWEEN ? AND ?
2944 WHERE MATCH(field) AGAINST (?, ?)
2946 Special operators IN and BETWEEN are fairly standard and therefore
2947 are builtin within C<SQL::Abstract> (as the overridable methods
2948 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2949 like the MATCH .. AGAINST example above which is specific to MySQL,
2950 you can write your own operator handlers - supply a C<special_ops>
2951 argument to the C<new> method. That argument takes an arrayref of
2952 operator definitions; each operator definition is a hashref with two
2959 the regular expression to match the operator
2963 Either a coderef or a plain scalar method name. In both cases
2964 the expected return is C<< ($sql, @bind) >>.
2966 When supplied with a method name, it is simply called on the
2967 L<SQL::Abstract> object as:
2969 $self->$method_name ($field, $op, $arg)
2973 $field is the LHS of the operator
2974 $op is the part that matched the handler regex
2977 When supplied with a coderef, it is called as:
2979 $coderef->($self, $field, $op, $arg)
2984 For example, here is an implementation
2985 of the MATCH .. AGAINST syntax for MySQL
2987 my $sqlmaker = SQL::Abstract->new(special_ops => [
2989 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2990 {regex => qr/^match$/i,
2992 my ($self, $field, $op, $arg) = @_;
2993 $arg = [$arg] if not ref $arg;
2994 my $label = $self->_quote($field);
2995 my ($placeholder) = $self->_convert('?');
2996 my $placeholders = join ", ", (($placeholder) x @$arg);
2997 my $sql = $self->_sqlcase('match') . " ($label) "
2998 . $self->_sqlcase('against') . " ($placeholders) ";
2999 my @bind = $self->_bindtype($field, @$arg);
3000 return ($sql, @bind);
3007 =head1 UNARY OPERATORS
3009 my $sqlmaker = SQL::Abstract->new(unary_ops => [
3013 my ($self, $op, $arg) = @_;
3019 handler => 'method_name',
3023 A "unary operator" is a SQL syntactic clause that can be
3024 applied to a field - the operator goes before the field
3026 You can write your own operator handlers - supply a C<unary_ops>
3027 argument to the C<new> method. That argument takes an arrayref of
3028 operator definitions; each operator definition is a hashref with two
3035 the regular expression to match the operator
3039 Either a coderef or a plain scalar method name. In both cases
3040 the expected return is C<< $sql >>.
3042 When supplied with a method name, it is simply called on the
3043 L<SQL::Abstract> object as:
3045 $self->$method_name ($op, $arg)
3049 $op is the part that matched the handler regex
3050 $arg is the RHS or argument of the operator
3052 When supplied with a coderef, it is called as:
3054 $coderef->($self, $op, $arg)
3062 Thanks to some benchmarking by Mark Stosberg, it turns out that
3063 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3064 I must admit this wasn't an intentional design issue, but it's a
3065 byproduct of the fact that you get to control your C<DBI> handles
3068 To maximize performance, use a code snippet like the following:
3070 # prepare a statement handle using the first row
3071 # and then reuse it for the rest of the rows
3073 for my $href (@array_of_hashrefs) {
3074 $stmt ||= $sql->insert('table', $href);
3075 $sth ||= $dbh->prepare($stmt);
3076 $sth->execute($sql->values($href));
3079 The reason this works is because the keys in your C<$href> are sorted
3080 internally by B<SQL::Abstract>. Thus, as long as your data retains
3081 the same structure, you only have to generate the SQL the first time
3082 around. On subsequent queries, simply use the C<values> function provided
3083 by this module to return your values in the correct order.
3085 However this depends on the values having the same type - if, for
3086 example, the values of a where clause may either have values
3087 (resulting in sql of the form C<column = ?> with a single bind
3088 value), or alternatively the values might be C<undef> (resulting in
3089 sql of the form C<column IS NULL> with no bind value) then the
3090 caching technique suggested will not work.
3094 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3095 really like this part (I do, at least). Building up a complex query
3096 can be as simple as the following:
3103 use CGI::FormBuilder;
3106 my $form = CGI::FormBuilder->new(...);
3107 my $sql = SQL::Abstract->new;
3109 if ($form->submitted) {
3110 my $field = $form->field;
3111 my $id = delete $field->{id};
3112 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3115 Of course, you would still have to connect using C<DBI> to run the
3116 query, but the point is that if you make your form look like your
3117 table, the actual query script can be extremely simplistic.
3119 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3120 a fast interface to returning and formatting data. I frequently
3121 use these three modules together to write complex database query
3122 apps in under 50 lines.
3124 =head1 HOW TO CONTRIBUTE
3126 Contributions are always welcome, in all usable forms (we especially
3127 welcome documentation improvements). The delivery methods include git-
3128 or unified-diff formatted patches, GitHub pull requests, or plain bug
3129 reports either via RT or the Mailing list. Contributors are generally
3130 granted full access to the official repository after their first several
3131 patches pass successful review.
3133 This project is maintained in a git repository. The code and related tools are
3134 accessible at the following locations:
3138 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3140 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3142 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3144 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3150 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3151 Great care has been taken to preserve the I<published> behavior
3152 documented in previous versions in the 1.* family; however,
3153 some features that were previously undocumented, or behaved
3154 differently from the documentation, had to be changed in order
3155 to clarify the semantics. Hence, client code that was relying
3156 on some dark areas of C<SQL::Abstract> v1.*
3157 B<might behave differently> in v1.50.
3159 The main changes are :
3165 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3169 support for the { operator => \"..." } construct (to embed literal SQL)
3173 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3177 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3181 defensive programming : check arguments
3185 fixed bug with global logic, which was previously implemented
3186 through global variables yielding side-effects. Prior versions would
3187 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3188 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3189 Now this is interpreted
3190 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3195 fixed semantics of _bindtype on array args
3199 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3200 we just avoid shifting arrays within that tree.
3204 dropped the C<_modlogic> function
3208 =head1 ACKNOWLEDGEMENTS
3210 There are a number of individuals that have really helped out with
3211 this module. Unfortunately, most of them submitted bugs via CPAN
3212 so I have no idea who they are! But the people I do know are:
3214 Ash Berlin (order_by hash term support)
3215 Matt Trout (DBIx::Class support)
3216 Mark Stosberg (benchmarking)
3217 Chas Owens (initial "IN" operator support)
3218 Philip Collins (per-field SQL functions)
3219 Eric Kolve (hashref "AND" support)
3220 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3221 Dan Kubb (support for "quote_char" and "name_sep")
3222 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3223 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3224 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3225 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3226 Oliver Charles (support for "RETURNING" after "INSERT")
3232 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3236 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3238 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3240 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3241 While not an official support venue, C<DBIx::Class> makes heavy use of
3242 C<SQL::Abstract>, and as such list members there are very familiar with
3243 how to create queries.
3247 This module is free software; you may copy this under the same
3248 terms as perl itself (either the GNU General Public License or
3249 the Artistic License)