1 package SQL::Abstract; # see doc at end of file
9 #======================================================================
11 #======================================================================
13 our $VERSION = '1.74';
15 # This would confuse some packagers
16 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
20 # special operators (-in, -between). May be extended/overridden by user.
21 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
22 my @BUILTIN_SPECIAL_OPS = (
23 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
24 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
25 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
26 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
27 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
30 # unaryish operators - key maps to handler
31 my @BUILTIN_UNARY_OPS = (
32 # the digits are backcompat stuff
33 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
34 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
35 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
36 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
37 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
38 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
41 #======================================================================
42 # DEBUGGING AND ERROR REPORTING
43 #======================================================================
46 return unless $_[0]->{debug}; shift; # a little faster
47 my $func = (caller(1))[3];
48 warn "[$func] ", @_, "\n";
52 my($func) = (caller(1))[3];
53 Carp::carp "[$func] Warning: ", @_;
57 my($func) = (caller(1))[3];
58 Carp::croak "[$func] Fatal: ", @_;
62 #======================================================================
64 #======================================================================
68 my $class = ref($self) || $self;
69 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
71 # choose our case by keeping an option around
72 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
74 # default logic for interpreting arrayrefs
75 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
77 # how to return bind vars
78 $opt{bindtype} ||= 'normal';
80 # default comparison is "=", but can be overridden
83 # try to recognize which are the 'equality' and 'inequality' ops
84 # (temporary quickfix (in 2007), should go through a more seasoned API)
85 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
86 $opt{inequality_op} = qr/^( != | <> )$/ix;
88 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
89 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
92 $opt{sqltrue} ||= '1=1';
93 $opt{sqlfalse} ||= '0=1';
96 $opt{special_ops} ||= [];
97 # regexes are applied in order, thus push after user-defines
98 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
101 $opt{unary_ops} ||= [];
102 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
104 # rudimentary sanity-check for user supplied bits treated as functions/operators
105 # If a purported function matches this regular expression, an exception is thrown.
106 # Literal SQL is *NOT* subject to this check, only functions (and column names
107 # when quoting is not in effect)
110 # need to guard against ()'s in column names too, but this will break tons of
111 # hacks... ideas anyone?
112 $opt{injection_guard} ||= qr/
118 return bless \%opt, $class;
122 sub _assert_pass_injection_guard {
123 if ($_[1] =~ $_[0]->{injection_guard}) {
124 my $class = ref $_[0];
125 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
126 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
127 . "{injection_guard} attribute to ${class}->new()"
132 #======================================================================
134 #======================================================================
138 my $table = $self->_table(shift);
139 my $data = shift || return;
142 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
143 my ($sql, @bind) = $self->$method($data);
144 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
146 if ($options->{returning}) {
147 my ($s, @b) = $self->_insert_returning ($options);
152 return wantarray ? ($sql, @bind) : $sql;
155 sub _insert_returning {
156 my ($self, $options) = @_;
158 my $f = $options->{returning};
160 my $fieldlist = $self->_SWITCH_refkind($f, {
161 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
162 SCALAR => sub {$self->_quote($f)},
163 SCALARREF => sub {$$f},
165 return $self->_sqlcase(' returning ') . $fieldlist;
168 sub _insert_HASHREF { # explicit list of fields and then values
169 my ($self, $data) = @_;
171 my @fields = sort keys %$data;
173 my ($sql, @bind) = $self->_insert_values($data);
176 $_ = $self->_quote($_) foreach @fields;
177 $sql = "( ".join(", ", @fields).") ".$sql;
179 return ($sql, @bind);
182 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
183 my ($self, $data) = @_;
185 # no names (arrayref) so can't generate bindtype
186 $self->{bindtype} ne 'columns'
187 or belch "can't do 'columns' bindtype when called with arrayref";
189 # fold the list of values into a hash of column name - value pairs
190 # (where the column names are artificially generated, and their
191 # lexicographical ordering keep the ordering of the original list)
192 my $i = "a"; # incremented values will be in lexicographical order
193 my $data_in_hash = { map { ($i++ => $_) } @$data };
195 return $self->_insert_values($data_in_hash);
198 sub _insert_ARRAYREFREF { # literal SQL with bind
199 my ($self, $data) = @_;
201 my ($sql, @bind) = @${$data};
202 $self->_assert_bindval_matches_bindtype(@bind);
204 return ($sql, @bind);
208 sub _insert_SCALARREF { # literal SQL without bind
209 my ($self, $data) = @_;
215 my ($self, $data) = @_;
217 my (@values, @all_bind);
218 foreach my $column (sort keys %$data) {
219 my $v = $data->{$column};
221 $self->_SWITCH_refkind($v, {
224 if ($self->{array_datatypes}) { # if array datatype are activated
226 push @all_bind, $self->_bindtype($column, $v);
228 else { # else literal SQL with bind
229 my ($sql, @bind) = @$v;
230 $self->_assert_bindval_matches_bindtype(@bind);
232 push @all_bind, @bind;
236 ARRAYREFREF => sub { # literal SQL with bind
237 my ($sql, @bind) = @${$v};
238 $self->_assert_bindval_matches_bindtype(@bind);
240 push @all_bind, @bind;
243 # THINK : anything useful to do with a HASHREF ?
244 HASHREF => sub { # (nothing, but old SQLA passed it through)
245 #TODO in SQLA >= 2.0 it will die instead
246 belch "HASH ref as bind value in insert is not supported";
248 push @all_bind, $self->_bindtype($column, $v);
251 SCALARREF => sub { # literal SQL without bind
255 SCALAR_or_UNDEF => sub {
257 push @all_bind, $self->_bindtype($column, $v);
264 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
265 return ($sql, @all_bind);
270 #======================================================================
272 #======================================================================
277 my $table = $self->_table(shift);
278 my $data = shift || return;
281 # first build the 'SET' part of the sql statement
282 my (@set, @all_bind);
283 puke "Unsupported data type specified to \$sql->update"
284 unless ref $data eq 'HASH';
286 for my $k (sort keys %$data) {
289 my $label = $self->_quote($k);
291 $self->_SWITCH_refkind($v, {
293 if ($self->{array_datatypes}) { # array datatype
294 push @set, "$label = ?";
295 push @all_bind, $self->_bindtype($k, $v);
297 else { # literal SQL with bind
298 my ($sql, @bind) = @$v;
299 $self->_assert_bindval_matches_bindtype(@bind);
300 push @set, "$label = $sql";
301 push @all_bind, @bind;
304 ARRAYREFREF => sub { # literal SQL with bind
305 my ($sql, @bind) = @${$v};
306 $self->_assert_bindval_matches_bindtype(@bind);
307 push @set, "$label = $sql";
308 push @all_bind, @bind;
310 SCALARREF => sub { # literal SQL without bind
311 push @set, "$label = $$v";
314 my ($op, $arg, @rest) = %$v;
316 puke 'Operator calls in update must be in the form { -op => $arg }'
317 if (@rest or not $op =~ /^\-(.+)/);
319 local $self->{_nested_func_lhs} = $k;
320 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
322 push @set, "$label = $sql";
323 push @all_bind, @bind;
325 SCALAR_or_UNDEF => sub {
326 push @set, "$label = ?";
327 push @all_bind, $self->_bindtype($k, $v);
333 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
337 my($where_sql, @where_bind) = $self->where($where);
339 push @all_bind, @where_bind;
342 return wantarray ? ($sql, @all_bind) : $sql;
348 #======================================================================
350 #======================================================================
355 my $table = $self->_table(shift);
356 my $fields = shift || '*';
360 my($where_sql, @bind) = $self->where($where, $order);
362 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
364 my $sql = join(' ', $self->_sqlcase('select'), $f,
365 $self->_sqlcase('from'), $table)
368 return wantarray ? ($sql, @bind) : $sql;
371 #======================================================================
373 #======================================================================
378 my $table = $self->_table(shift);
382 my($where_sql, @bind) = $self->where($where);
383 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
385 return wantarray ? ($sql, @bind) : $sql;
389 #======================================================================
391 #======================================================================
395 # Finally, a separate routine just to handle WHERE clauses
397 my ($self, $where, $order) = @_;
400 my ($sql, @bind) = $self->_recurse_where($where);
401 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
405 $sql .= $self->_order_by($order);
408 return wantarray ? ($sql, @bind) : $sql;
413 my ($self, $where, $logic) = @_;
415 # dispatch on appropriate method according to refkind of $where
416 my $method = $self->_METHOD_FOR_refkind("_where", $where);
418 my ($sql, @bind) = $self->$method($where, $logic);
420 # DBIx::Class directly calls _recurse_where in scalar context, so
421 # we must implement it, even if not in the official API
422 return wantarray ? ($sql, @bind) : $sql;
427 #======================================================================
428 # WHERE: top-level ARRAYREF
429 #======================================================================
432 sub _where_ARRAYREF {
433 my ($self, $where, $logic) = @_;
435 $logic = uc($logic || $self->{logic});
436 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
438 my @clauses = @$where;
440 my (@sql_clauses, @all_bind);
441 # need to use while() so can shift() for pairs
442 while (my $el = shift @clauses) {
444 # switch according to kind of $el and get corresponding ($sql, @bind)
445 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
447 # skip empty elements, otherwise get invalid trailing AND stuff
448 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
452 $self->_assert_bindval_matches_bindtype(@b);
456 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
458 SCALARREF => sub { ($$el); },
460 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
461 $self->_recurse_where({$el => shift(@clauses)})},
463 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
467 push @sql_clauses, $sql;
468 push @all_bind, @bind;
472 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
475 #======================================================================
476 # WHERE: top-level ARRAYREFREF
477 #======================================================================
479 sub _where_ARRAYREFREF {
480 my ($self, $where) = @_;
481 my ($sql, @bind) = @$$where;
482 $self->_assert_bindval_matches_bindtype(@bind);
483 return ($sql, @bind);
486 #======================================================================
487 # WHERE: top-level HASHREF
488 #======================================================================
491 my ($self, $where) = @_;
492 my (@sql_clauses, @all_bind);
494 for my $k (sort keys %$where) {
495 my $v = $where->{$k};
497 # ($k => $v) is either a special unary op or a regular hashpair
498 my ($sql, @bind) = do {
500 # put the operator in canonical form
502 $op = substr $op, 1; # remove initial dash
503 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
504 $op =~ s/\s+/ /g; # compress whitespace
506 # so that -not_foo works correctly
507 $op =~ s/^not_/NOT /i;
509 $self->_debug("Unary OP(-$op) within hashref, recursing...");
510 my ($s, @b) = $self->_where_unary_op ($op, $v);
512 # top level vs nested
513 # we assume that handled unary ops will take care of their ()s
515 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
517 defined($self->{_nested_func_lhs}) && ($self->{_nested_func_lhs} eq $k)
522 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
523 $self->$method($k, $v);
527 push @sql_clauses, $sql;
528 push @all_bind, @bind;
531 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
534 sub _where_unary_op {
535 my ($self, $op, $rhs) = @_;
537 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
538 my $handler = $op_entry->{handler};
540 if (not ref $handler) {
541 if ($op =~ s/ [_\s]? \d+ $//x ) {
542 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
543 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
545 return $self->$handler ($op, $rhs);
547 elsif (ref $handler eq 'CODE') {
548 return $handler->($self, $op, $rhs);
551 puke "Illegal handler for operator $op - expecting a method name or a coderef";
555 $self->_debug("Generic unary OP: $op - recursing as function");
557 $self->_assert_pass_injection_guard($op);
559 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
561 puke "Illegal use of top-level '$op'"
562 unless $self->{_nested_func_lhs};
565 $self->_convert('?'),
566 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
570 $self->_recurse_where ($rhs)
574 $sql = sprintf ('%s %s',
575 $self->_sqlcase($op),
579 return ($sql, @bind);
582 sub _where_op_ANDOR {
583 my ($self, $op, $v) = @_;
585 $self->_SWITCH_refkind($v, {
587 return $self->_where_ARRAYREF($v, $op);
591 return ( $op =~ /^or/i )
592 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
593 : $self->_where_HASHREF($v);
597 puke "-$op => \\\$scalar makes little sense, use " .
599 ? '[ \$scalar, \%rest_of_conditions ] instead'
600 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
605 puke "-$op => \\[...] makes little sense, use " .
607 ? '[ \[...], \%rest_of_conditions ] instead'
608 : '-and => [ \[...], \%rest_of_conditions ] instead'
612 SCALAR => sub { # permissively interpreted as SQL
613 puke "-$op => \$value makes little sense, use -bool => \$value instead";
617 puke "-$op => undef not supported";
623 my ($self, $op, $v) = @_;
625 $self->_SWITCH_refkind($v, {
627 SCALAR => sub { # permissively interpreted as SQL
628 belch "literal SQL should be -nest => \\'scalar' "
629 . "instead of -nest => 'scalar' ";
634 puke "-$op => undef not supported";
638 $self->_recurse_where ($v);
646 my ($self, $op, $v) = @_;
648 my ($s, @b) = $self->_SWITCH_refkind($v, {
649 SCALAR => sub { # interpreted as SQL column
650 $self->_convert($self->_quote($v));
654 puke "-$op => undef not supported";
658 $self->_recurse_where ($v);
662 $s = "(NOT $s)" if $op =~ /^not/i;
667 sub _where_op_IDENT {
669 my ($op, $rhs) = splice @_, -2;
671 puke "-$op takes a single scalar argument (a quotable identifier)";
674 # in case we are called as a top level special op (no '=')
677 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
685 sub _where_op_VALUE {
687 my ($op, $rhs) = splice @_, -2;
689 # in case we are called as a top level special op (no '=')
694 ($lhs || $self->{_nested_func_lhs}),
701 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
705 $self->_convert('?'),
711 sub _where_hashpair_ARRAYREF {
712 my ($self, $k, $v) = @_;
715 my @v = @$v; # need copy because of shift below
716 $self->_debug("ARRAY($k) means distribute over elements");
718 # put apart first element if it is an operator (-and, -or)
720 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
724 my @distributed = map { {$k => $_} } @v;
727 $self->_debug("OP($op) reinjected into the distributed array");
728 unshift @distributed, $op;
731 my $logic = $op ? substr($op, 1) : '';
733 return $self->_recurse_where(\@distributed, $logic);
736 $self->_debug("empty ARRAY($k) means 0=1");
737 return ($self->{sqlfalse});
741 sub _where_hashpair_HASHREF {
742 my ($self, $k, $v, $logic) = @_;
745 local $self->{_nested_func_lhs} = $self->{_nested_func_lhs};
747 my ($all_sql, @all_bind);
749 for my $orig_op (sort keys %$v) {
750 my $val = $v->{$orig_op};
752 # put the operator in canonical form
755 # FIXME - we need to phase out dash-less ops
756 $op =~ s/^-//; # remove possible initial dash
757 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
758 $op =~ s/\s+/ /g; # compress whitespace
760 $self->_assert_pass_injection_guard($op);
763 $op =~ s/^is_not/IS NOT/i;
765 # so that -not_foo works correctly
766 $op =~ s/^not_/NOT /i;
770 # CASE: col-value logic modifiers
771 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
772 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
774 # CASE: special operators like -in or -between
775 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
776 my $handler = $special_op->{handler};
778 puke "No handler supplied for special operator $orig_op";
780 elsif (not ref $handler) {
781 ($sql, @bind) = $self->$handler ($k, $op, $val);
783 elsif (ref $handler eq 'CODE') {
784 ($sql, @bind) = $handler->($self, $k, $op, $val);
787 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
791 $self->_SWITCH_refkind($val, {
793 ARRAYREF => sub { # CASE: col => {op => \@vals}
794 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
797 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
798 my ($sub_sql, @sub_bind) = @$$val;
799 $self->_assert_bindval_matches_bindtype(@sub_bind);
800 $sql = join ' ', $self->_convert($self->_quote($k)),
801 $self->_sqlcase($op),
806 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
808 $op =~ $self->{equality_op} ? 'is'
809 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
810 : $op =~ $self->{inequality_op} ? 'is not'
811 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
812 : puke "unexpected operator '$orig_op' with undef operand";
814 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
817 FALLBACK => sub { # CASE: col => {op/func => $stuff}
819 # retain for proper column type bind
820 $self->{_nested_func_lhs} ||= $k;
822 ($sql, @bind) = $self->_where_unary_op ($op, $val);
825 $self->_convert($self->_quote($k)),
826 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
832 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
833 push @all_bind, @bind;
835 return ($all_sql, @all_bind);
838 sub _where_field_IS {
839 my ($self, $k, $op, $v) = @_;
841 my ($s) = $self->_SWITCH_refkind($v, {
844 $self->_convert($self->_quote($k)),
845 map { $self->_sqlcase($_)} ($op, 'null')
848 puke "$op can only take undef as argument";
855 sub _where_field_op_ARRAYREF {
856 my ($self, $k, $op, $vals) = @_;
858 my @vals = @$vals; #always work on a copy
861 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
863 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
866 # see if the first element is an -and/-or op
868 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
873 # a long standing API wart - an attempt to change this behavior during
874 # the 1.50 series failed *spectacularly*. Warn instead and leave the
879 (!$logic or $logic eq 'OR')
881 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
884 belch "A multi-element arrayref as an argument to the inequality op '$o' "
885 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
886 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
890 # distribute $op over each remaining member of @vals, append logic if exists
891 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
895 # try to DWIM on equality operators
897 $op =~ $self->{equality_op} ? $self->{sqlfalse}
898 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
899 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
900 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
901 : puke "operator '$op' applied on an empty array (field '$k')";
906 sub _where_hashpair_SCALARREF {
907 my ($self, $k, $v) = @_;
908 $self->_debug("SCALAR($k) means literal SQL: $$v");
909 my $sql = $self->_quote($k) . " " . $$v;
913 # literal SQL with bind
914 sub _where_hashpair_ARRAYREFREF {
915 my ($self, $k, $v) = @_;
916 $self->_debug("REF($k) means literal SQL: @${$v}");
917 my ($sql, @bind) = @$$v;
918 $self->_assert_bindval_matches_bindtype(@bind);
919 $sql = $self->_quote($k) . " " . $sql;
920 return ($sql, @bind );
923 # literal SQL without bind
924 sub _where_hashpair_SCALAR {
925 my ($self, $k, $v) = @_;
926 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
927 my $sql = join ' ', $self->_convert($self->_quote($k)),
928 $self->_sqlcase($self->{cmp}),
929 $self->_convert('?');
930 my @bind = $self->_bindtype($k, $v);
931 return ( $sql, @bind);
935 sub _where_hashpair_UNDEF {
936 my ($self, $k, $v) = @_;
937 $self->_debug("UNDEF($k) means IS NULL");
938 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
942 #======================================================================
943 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
944 #======================================================================
947 sub _where_SCALARREF {
948 my ($self, $where) = @_;
951 $self->_debug("SCALAR(*top) means literal SQL: $$where");
957 my ($self, $where) = @_;
960 $self->_debug("NOREF(*top) means literal SQL: $where");
971 #======================================================================
972 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
973 #======================================================================
976 sub _where_field_BETWEEN {
977 my ($self, $k, $op, $vals) = @_;
979 my ($label, $and, $placeholder);
980 $label = $self->_convert($self->_quote($k));
981 $and = ' ' . $self->_sqlcase('and') . ' ';
982 $placeholder = $self->_convert('?');
983 $op = $self->_sqlcase($op);
985 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
987 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
989 my ($s, @b) = @$$vals;
990 $self->_assert_bindval_matches_bindtype(@b);
997 puke $invalid_args if @$vals != 2;
999 my (@all_sql, @all_bind);
1000 foreach my $val (@$vals) {
1001 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1003 return ($placeholder, $self->_bindtype($k, $val) );
1008 ARRAYREFREF => sub {
1009 my ($sql, @bind) = @$$val;
1010 $self->_assert_bindval_matches_bindtype(@bind);
1011 return ($sql, @bind);
1014 my ($func, $arg, @rest) = %$val;
1015 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1016 if (@rest or $func !~ /^ \- (.+)/x);
1017 local $self->{_nested_func_lhs} = $k;
1018 $self->_where_unary_op ($1 => $arg);
1024 push @all_sql, $sql;
1025 push @all_bind, @bind;
1029 (join $and, @all_sql),
1038 my $sql = "( $label $op $clause )";
1039 return ($sql, @bind)
1043 sub _where_field_IN {
1044 my ($self, $k, $op, $vals) = @_;
1046 # backwards compatibility : if scalar, force into an arrayref
1047 $vals = [$vals] if defined $vals && ! ref $vals;
1049 my ($label) = $self->_convert($self->_quote($k));
1050 my ($placeholder) = $self->_convert('?');
1051 $op = $self->_sqlcase($op);
1053 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1054 ARRAYREF => sub { # list of choices
1055 if (@$vals) { # nonempty list
1056 my (@all_sql, @all_bind);
1058 for my $val (@$vals) {
1059 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1061 return ($placeholder, $val);
1066 ARRAYREFREF => sub {
1067 my ($sql, @bind) = @$$val;
1068 $self->_assert_bindval_matches_bindtype(@bind);
1069 return ($sql, @bind);
1072 my ($func, $arg, @rest) = %$val;
1073 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1074 if (@rest or $func !~ /^ \- (.+)/x);
1075 local $self->{_nested_func_lhs} = $k;
1076 $self->_where_unary_op ($1 => $arg);
1080 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1081 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1082 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1083 . 'will emit the logically correct SQL instead of raising this exception)'
1087 push @all_sql, $sql;
1088 push @all_bind, @bind;
1092 sprintf ('%s %s ( %s )',
1095 join (', ', @all_sql)
1097 $self->_bindtype($k, @all_bind),
1100 else { # empty list : some databases won't understand "IN ()", so DWIM
1101 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1106 SCALARREF => sub { # literal SQL
1107 my $sql = $self->_open_outer_paren ($$vals);
1108 return ("$label $op ( $sql )");
1110 ARRAYREFREF => sub { # literal SQL with bind
1111 my ($sql, @bind) = @$$vals;
1112 $self->_assert_bindval_matches_bindtype(@bind);
1113 $sql = $self->_open_outer_paren ($sql);
1114 return ("$label $op ( $sql )", @bind);
1118 puke "Argument passed to the '$op' operator can not be undefined";
1122 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1126 return ($sql, @bind);
1129 # Some databases (SQLite) treat col IN (1, 2) different from
1130 # col IN ( (1, 2) ). Use this to strip all outer parens while
1131 # adding them back in the corresponding method
1132 sub _open_outer_paren {
1133 my ($self, $sql) = @_;
1134 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/xs;
1139 #======================================================================
1141 #======================================================================
1144 my ($self, $arg) = @_;
1147 for my $c ($self->_order_by_chunks ($arg) ) {
1148 $self->_SWITCH_refkind ($c, {
1149 SCALAR => sub { push @sql, $c },
1150 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1156 $self->_sqlcase(' order by'),
1162 return wantarray ? ($sql, @bind) : $sql;
1165 sub _order_by_chunks {
1166 my ($self, $arg) = @_;
1168 return $self->_SWITCH_refkind($arg, {
1171 map { $self->_order_by_chunks ($_ ) } @$arg;
1174 ARRAYREFREF => sub {
1175 my ($s, @b) = @$$arg;
1176 $self->_assert_bindval_matches_bindtype(@b);
1180 SCALAR => sub {$self->_quote($arg)},
1182 UNDEF => sub {return () },
1184 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1187 # get first pair in hash
1188 my ($key, $val, @rest) = %$arg;
1190 return () unless $key;
1192 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1193 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1199 for my $c ($self->_order_by_chunks ($val)) {
1202 $self->_SWITCH_refkind ($c, {
1207 ($sql, @bind) = @$c;
1211 $sql = $sql . ' ' . $self->_sqlcase($direction);
1213 push @ret, [ $sql, @bind];
1222 #======================================================================
1223 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1224 #======================================================================
1229 $self->_SWITCH_refkind($from, {
1230 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1231 SCALAR => sub {$self->_quote($from)},
1232 SCALARREF => sub {$$from},
1237 #======================================================================
1239 #======================================================================
1241 # highly optimized, as it's called way too often
1243 # my ($self, $label) = @_;
1245 return '' unless defined $_[1];
1246 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1248 unless ($_[0]->{quote_char}) {
1249 $_[0]->_assert_pass_injection_guard($_[1]);
1253 my $qref = ref $_[0]->{quote_char};
1256 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1258 elsif ($qref eq 'ARRAY') {
1259 ($l, $r) = @{$_[0]->{quote_char}};
1262 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1265 # parts containing * are naturally unquoted
1266 return join( $_[0]->{name_sep}||'', map
1267 { $_ eq '*' ? $_ : $l . $_ . $r }
1268 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1273 # Conversion, if applicable
1275 #my ($self, $arg) = @_;
1276 if ($_[0]->{convert}) {
1277 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1284 #my ($self, $col, @vals) = @_;
1285 # called often - tighten code
1286 return $_[0]->{bindtype} eq 'columns'
1287 ? map {[$_[1], $_]} @_[2 .. $#_]
1292 # Dies if any element of @bind is not in [colname => value] format
1293 # if bindtype is 'columns'.
1294 sub _assert_bindval_matches_bindtype {
1295 # my ($self, @bind) = @_;
1297 if ($self->{bindtype} eq 'columns') {
1299 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1300 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1306 sub _join_sql_clauses {
1307 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1309 if (@$clauses_aref > 1) {
1310 my $join = " " . $self->_sqlcase($logic) . " ";
1311 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1312 return ($sql, @$bind_aref);
1314 elsif (@$clauses_aref) {
1315 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1318 return (); # if no SQL, ignore @$bind_aref
1323 # Fix SQL case, if so requested
1325 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1326 # don't touch the argument ... crooked logic, but let's not change it!
1327 return $_[0]->{case} ? $_[1] : uc($_[1]);
1331 #======================================================================
1332 # DISPATCHING FROM REFKIND
1333 #======================================================================
1336 my ($self, $data) = @_;
1338 return 'UNDEF' unless defined $data;
1340 # blessed objects are treated like scalars
1341 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1343 return 'SCALAR' unless $ref;
1346 while ($ref eq 'REF') {
1348 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1352 return ($ref||'SCALAR') . ('REF' x $n_steps);
1356 my ($self, $data) = @_;
1357 my @try = ($self->_refkind($data));
1358 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1359 push @try, 'FALLBACK';
1363 sub _METHOD_FOR_refkind {
1364 my ($self, $meth_prefix, $data) = @_;
1367 for (@{$self->_try_refkind($data)}) {
1368 $method = $self->can($meth_prefix."_".$_)
1372 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1376 sub _SWITCH_refkind {
1377 my ($self, $data, $dispatch_table) = @_;
1380 for (@{$self->_try_refkind($data)}) {
1381 $coderef = $dispatch_table->{$_}
1385 puke "no dispatch entry for ".$self->_refkind($data)
1394 #======================================================================
1395 # VALUES, GENERATE, AUTOLOAD
1396 #======================================================================
1398 # LDNOTE: original code from nwiger, didn't touch code in that section
1399 # I feel the AUTOLOAD stuff should not be the default, it should
1400 # only be activated on explicit demand by user.
1404 my $data = shift || return;
1405 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1406 unless ref $data eq 'HASH';
1409 foreach my $k ( sort keys %$data ) {
1410 my $v = $data->{$k};
1411 $self->_SWITCH_refkind($v, {
1413 if ($self->{array_datatypes}) { # array datatype
1414 push @all_bind, $self->_bindtype($k, $v);
1416 else { # literal SQL with bind
1417 my ($sql, @bind) = @$v;
1418 $self->_assert_bindval_matches_bindtype(@bind);
1419 push @all_bind, @bind;
1422 ARRAYREFREF => sub { # literal SQL with bind
1423 my ($sql, @bind) = @${$v};
1424 $self->_assert_bindval_matches_bindtype(@bind);
1425 push @all_bind, @bind;
1427 SCALARREF => sub { # literal SQL without bind
1429 SCALAR_or_UNDEF => sub {
1430 push @all_bind, $self->_bindtype($k, $v);
1441 my(@sql, @sqlq, @sqlv);
1445 if ($ref eq 'HASH') {
1446 for my $k (sort keys %$_) {
1449 my $label = $self->_quote($k);
1450 if ($r eq 'ARRAY') {
1451 # literal SQL with bind
1452 my ($sql, @bind) = @$v;
1453 $self->_assert_bindval_matches_bindtype(@bind);
1454 push @sqlq, "$label = $sql";
1456 } elsif ($r eq 'SCALAR') {
1457 # literal SQL without bind
1458 push @sqlq, "$label = $$v";
1460 push @sqlq, "$label = ?";
1461 push @sqlv, $self->_bindtype($k, $v);
1464 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1465 } elsif ($ref eq 'ARRAY') {
1466 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1469 if ($r eq 'ARRAY') { # literal SQL with bind
1470 my ($sql, @bind) = @$v;
1471 $self->_assert_bindval_matches_bindtype(@bind);
1474 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1475 # embedded literal SQL
1482 push @sql, '(' . join(', ', @sqlq) . ')';
1483 } elsif ($ref eq 'SCALAR') {
1487 # strings get case twiddled
1488 push @sql, $self->_sqlcase($_);
1492 my $sql = join ' ', @sql;
1494 # this is pretty tricky
1495 # if ask for an array, return ($stmt, @bind)
1496 # otherwise, s/?/shift @sqlv/ to put it inline
1498 return ($sql, @sqlv);
1500 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1501 ref $d ? $d->[1] : $d/e;
1510 # This allows us to check for a local, then _form, attr
1512 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1513 return $self->generate($name, @_);
1524 SQL::Abstract - Generate SQL from Perl data structures
1530 my $sql = SQL::Abstract->new;
1532 my($stmt, @bind) = $sql->select($source, \@fields, \%where, \@order);
1534 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1536 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1538 my($stmt, @bind) = $sql->delete($table, \%where);
1540 # Then, use these in your DBI statements
1541 my $sth = $dbh->prepare($stmt);
1542 $sth->execute(@bind);
1544 # Just generate the WHERE clause
1545 my($stmt, @bind) = $sql->where(\%where, \@order);
1547 # Return values in the same order, for hashed queries
1548 # See PERFORMANCE section for more details
1549 my @bind = $sql->values(\%fieldvals);
1553 This module was inspired by the excellent L<DBIx::Abstract>.
1554 However, in using that module I found that what I really wanted
1555 to do was generate SQL, but still retain complete control over my
1556 statement handles and use the DBI interface. So, I set out to
1557 create an abstract SQL generation module.
1559 While based on the concepts used by L<DBIx::Abstract>, there are
1560 several important differences, especially when it comes to WHERE
1561 clauses. I have modified the concepts used to make the SQL easier
1562 to generate from Perl data structures and, IMO, more intuitive.
1563 The underlying idea is for this module to do what you mean, based
1564 on the data structures you provide it. The big advantage is that
1565 you don't have to modify your code every time your data changes,
1566 as this module figures it out.
1568 To begin with, an SQL INSERT is as easy as just specifying a hash
1569 of C<key=value> pairs:
1572 name => 'Jimbo Bobson',
1573 phone => '123-456-7890',
1574 address => '42 Sister Lane',
1575 city => 'St. Louis',
1576 state => 'Louisiana',
1579 The SQL can then be generated with this:
1581 my($stmt, @bind) = $sql->insert('people', \%data);
1583 Which would give you something like this:
1585 $stmt = "INSERT INTO people
1586 (address, city, name, phone, state)
1587 VALUES (?, ?, ?, ?, ?)";
1588 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1589 '123-456-7890', 'Louisiana');
1591 These are then used directly in your DBI code:
1593 my $sth = $dbh->prepare($stmt);
1594 $sth->execute(@bind);
1596 =head2 Inserting and Updating Arrays
1598 If your database has array types (like for example Postgres),
1599 activate the special option C<< array_datatypes => 1 >>
1600 when creating the C<SQL::Abstract> object.
1601 Then you may use an arrayref to insert and update database array types:
1603 my $sql = SQL::Abstract->new(array_datatypes => 1);
1605 planets => [qw/Mercury Venus Earth Mars/]
1608 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1612 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1614 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1617 =head2 Inserting and Updating SQL
1619 In order to apply SQL functions to elements of your C<%data> you may
1620 specify a reference to an arrayref for the given hash value. For example,
1621 if you need to execute the Oracle C<to_date> function on a value, you can
1622 say something like this:
1626 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1629 The first value in the array is the actual SQL. Any other values are
1630 optional and would be included in the bind values array. This gives
1633 my($stmt, @bind) = $sql->insert('people', \%data);
1635 $stmt = "INSERT INTO people (name, date_entered)
1636 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1637 @bind = ('Bill', '03/02/2003');
1639 An UPDATE is just as easy, all you change is the name of the function:
1641 my($stmt, @bind) = $sql->update('people', \%data);
1643 Notice that your C<%data> isn't touched; the module will generate
1644 the appropriately quirky SQL for you automatically. Usually you'll
1645 want to specify a WHERE clause for your UPDATE, though, which is
1646 where handling C<%where> hashes comes in handy...
1648 =head2 Complex where statements
1650 This module can generate pretty complicated WHERE statements
1651 easily. For example, simple C<key=value> pairs are taken to mean
1652 equality, and if you want to see if a field is within a set
1653 of values, you can use an arrayref. Let's say we wanted to
1654 SELECT some data based on this criteria:
1657 requestor => 'inna',
1658 worker => ['nwiger', 'rcwe', 'sfz'],
1659 status => { '!=', 'completed' }
1662 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1664 The above would give you something like this:
1666 $stmt = "SELECT * FROM tickets WHERE
1667 ( requestor = ? ) AND ( status != ? )
1668 AND ( worker = ? OR worker = ? OR worker = ? )";
1669 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1671 Which you could then use in DBI code like so:
1673 my $sth = $dbh->prepare($stmt);
1674 $sth->execute(@bind);
1680 The functions are simple. There's one for each major SQL operation,
1681 and a constructor you use first. The arguments are specified in a
1682 similar order to each function (table, then fields, then a where
1683 clause) to try and simplify things.
1688 =head2 new(option => 'value')
1690 The C<new()> function takes a list of options and values, and returns
1691 a new B<SQL::Abstract> object which can then be used to generate SQL
1692 through the methods below. The options accepted are:
1698 If set to 'lower', then SQL will be generated in all lowercase. By
1699 default SQL is generated in "textbook" case meaning something like:
1701 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1703 Any setting other than 'lower' is ignored.
1707 This determines what the default comparison operator is. By default
1708 it is C<=>, meaning that a hash like this:
1710 %where = (name => 'nwiger', email => 'nate@wiger.org');
1712 Will generate SQL like this:
1714 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1716 However, you may want loose comparisons by default, so if you set
1717 C<cmp> to C<like> you would get SQL such as:
1719 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1721 You can also override the comparison on an individual basis - see
1722 the huge section on L</"WHERE CLAUSES"> at the bottom.
1724 =item sqltrue, sqlfalse
1726 Expressions for inserting boolean values within SQL statements.
1727 By default these are C<1=1> and C<1=0>. They are used
1728 by the special operators C<-in> and C<-not_in> for generating
1729 correct SQL even when the argument is an empty array (see below).
1733 This determines the default logical operator for multiple WHERE
1734 statements in arrays or hashes. If absent, the default logic is "or"
1735 for arrays, and "and" for hashes. This means that a WHERE
1739 event_date => {'>=', '2/13/99'},
1740 event_date => {'<=', '4/24/03'},
1743 will generate SQL like this:
1745 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1747 This is probably not what you want given this query, though (look
1748 at the dates). To change the "OR" to an "AND", simply specify:
1750 my $sql = SQL::Abstract->new(logic => 'and');
1752 Which will change the above C<WHERE> to:
1754 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1756 The logic can also be changed locally by inserting
1757 a modifier in front of an arrayref :
1759 @where = (-and => [event_date => {'>=', '2/13/99'},
1760 event_date => {'<=', '4/24/03'} ]);
1762 See the L</"WHERE CLAUSES"> section for explanations.
1766 This will automatically convert comparisons using the specified SQL
1767 function for both column and value. This is mostly used with an argument
1768 of C<upper> or C<lower>, so that the SQL will have the effect of
1769 case-insensitive "searches". For example, this:
1771 $sql = SQL::Abstract->new(convert => 'upper');
1772 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1774 Will turn out the following SQL:
1776 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1778 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1779 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1780 not validate this option; it will just pass through what you specify verbatim).
1784 This is a kludge because many databases suck. For example, you can't
1785 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1786 Instead, you have to use C<bind_param()>:
1788 $sth->bind_param(1, 'reg data');
1789 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1791 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1792 which loses track of which field each slot refers to. Fear not.
1794 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1795 Currently, you can specify either C<normal> (default) or C<columns>. If you
1796 specify C<columns>, you will get an array that looks like this:
1798 my $sql = SQL::Abstract->new(bindtype => 'columns');
1799 my($stmt, @bind) = $sql->insert(...);
1802 [ 'column1', 'value1' ],
1803 [ 'column2', 'value2' ],
1804 [ 'column3', 'value3' ],
1807 You can then iterate through this manually, using DBI's C<bind_param()>.
1809 $sth->prepare($stmt);
1812 my($col, $data) = @$_;
1813 if ($col eq 'details' || $col eq 'comments') {
1814 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1815 } elsif ($col eq 'image') {
1816 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1818 $sth->bind_param($i, $data);
1822 $sth->execute; # execute without @bind now
1824 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1825 Basically, the advantage is still that you don't have to care which fields
1826 are or are not included. You could wrap that above C<for> loop in a simple
1827 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1828 get a layer of abstraction over manual SQL specification.
1830 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1831 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1832 will expect the bind values in this format.
1836 This is the character that a table or column name will be quoted
1837 with. By default this is an empty string, but you could set it to
1838 the character C<`>, to generate SQL like this:
1840 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1842 Alternatively, you can supply an array ref of two items, the first being the left
1843 hand quote character, and the second the right hand quote character. For
1844 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1845 that generates SQL like this:
1847 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1849 Quoting is useful if you have tables or columns names that are reserved
1850 words in your database's SQL dialect.
1854 This is the character that separates a table and column name. It is
1855 necessary to specify this when the C<quote_char> option is selected,
1856 so that tables and column names can be individually quoted like this:
1858 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1860 =item injection_guard
1862 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1863 column name specified in a query structure. This is a safety mechanism to avoid
1864 injection attacks when mishandling user input e.g.:
1866 my %condition_as_column_value_pairs = get_values_from_user();
1867 $sqla->select( ... , \%condition_as_column_value_pairs );
1869 If the expression matches an exception is thrown. Note that literal SQL
1870 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1872 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1874 =item array_datatypes
1876 When this option is true, arrayrefs in INSERT or UPDATE are
1877 interpreted as array datatypes and are passed directly
1879 When this option is false, arrayrefs are interpreted
1880 as literal SQL, just like refs to arrayrefs
1881 (but this behavior is for backwards compatibility; when writing
1882 new queries, use the "reference to arrayref" syntax
1888 Takes a reference to a list of "special operators"
1889 to extend the syntax understood by L<SQL::Abstract>.
1890 See section L</"SPECIAL OPERATORS"> for details.
1894 Takes a reference to a list of "unary operators"
1895 to extend the syntax understood by L<SQL::Abstract>.
1896 See section L</"UNARY OPERATORS"> for details.
1902 =head2 insert($table, \@values || \%fieldvals, \%options)
1904 This is the simplest function. You simply give it a table name
1905 and either an arrayref of values or hashref of field/value pairs.
1906 It returns an SQL INSERT statement and a list of bind values.
1907 See the sections on L</"Inserting and Updating Arrays"> and
1908 L</"Inserting and Updating SQL"> for information on how to insert
1909 with those data types.
1911 The optional C<\%options> hash reference may contain additional
1912 options to generate the insert SQL. Currently supported options
1919 Takes either a scalar of raw SQL fields, or an array reference of
1920 field names, and adds on an SQL C<RETURNING> statement at the end.
1921 This allows you to return data generated by the insert statement
1922 (such as row IDs) without performing another C<SELECT> statement.
1923 Note, however, this is not part of the SQL standard and may not
1924 be supported by all database engines.
1928 =head2 update($table, \%fieldvals, \%where)
1930 This takes a table, hashref of field/value pairs, and an optional
1931 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1933 See the sections on L</"Inserting and Updating Arrays"> and
1934 L</"Inserting and Updating SQL"> for information on how to insert
1935 with those data types.
1937 =head2 select($source, $fields, $where, $order)
1939 This returns a SQL SELECT statement and associated list of bind values, as
1940 specified by the arguments :
1946 Specification of the 'FROM' part of the statement.
1947 The argument can be either a plain scalar (interpreted as a table
1948 name, will be quoted), or an arrayref (interpreted as a list
1949 of table names, joined by commas, quoted), or a scalarref
1950 (literal table name, not quoted), or a ref to an arrayref
1951 (list of literal table names, joined by commas, not quoted).
1955 Specification of the list of fields to retrieve from
1957 The argument can be either an arrayref (interpreted as a list
1958 of field names, will be joined by commas and quoted), or a
1959 plain scalar (literal SQL, not quoted).
1960 Please observe that this API is not as flexible as that of
1961 the first argument C<$source>, for backwards compatibility reasons.
1965 Optional argument to specify the WHERE part of the query.
1966 The argument is most often a hashref, but can also be
1967 an arrayref or plain scalar --
1968 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1972 Optional argument to specify the ORDER BY part of the query.
1973 The argument can be a scalar, a hashref or an arrayref
1974 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1980 =head2 delete($table, \%where)
1982 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1983 It returns an SQL DELETE statement and list of bind values.
1985 =head2 where(\%where, \@order)
1987 This is used to generate just the WHERE clause. For example,
1988 if you have an arbitrary data structure and know what the
1989 rest of your SQL is going to look like, but want an easy way
1990 to produce a WHERE clause, use this. It returns an SQL WHERE
1991 clause and list of bind values.
1994 =head2 values(\%data)
1996 This just returns the values from the hash C<%data>, in the same
1997 order that would be returned from any of the other above queries.
1998 Using this allows you to markedly speed up your queries if you
1999 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2001 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2003 Warning: This is an experimental method and subject to change.
2005 This returns arbitrarily generated SQL. It's a really basic shortcut.
2006 It will return two different things, depending on return context:
2008 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2009 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2011 These would return the following:
2013 # First calling form
2014 $stmt = "CREATE TABLE test (?, ?)";
2015 @bind = (field1, field2);
2017 # Second calling form
2018 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2020 Depending on what you're trying to do, it's up to you to choose the correct
2021 format. In this example, the second form is what you would want.
2025 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2029 ALTER SESSION SET nls_date_format = 'MM/YY'
2031 You get the idea. Strings get their case twiddled, but everything
2032 else remains verbatim.
2034 =head1 WHERE CLAUSES
2038 This module uses a variation on the idea from L<DBIx::Abstract>. It
2039 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2040 module is that things in arrays are OR'ed, and things in hashes
2043 The easiest way to explain is to show lots of examples. After
2044 each C<%where> hash shown, it is assumed you used:
2046 my($stmt, @bind) = $sql->where(\%where);
2048 However, note that the C<%where> hash can be used directly in any
2049 of the other functions as well, as described above.
2051 =head2 Key-value pairs
2053 So, let's get started. To begin, a simple hash:
2057 status => 'completed'
2060 Is converted to SQL C<key = val> statements:
2062 $stmt = "WHERE user = ? AND status = ?";
2063 @bind = ('nwiger', 'completed');
2065 One common thing I end up doing is having a list of values that
2066 a field can be in. To do this, simply specify a list inside of
2071 status => ['assigned', 'in-progress', 'pending'];
2074 This simple code will create the following:
2076 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2077 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2079 A field associated to an empty arrayref will be considered a
2080 logical false and will generate 0=1.
2082 =head2 Tests for NULL values
2084 If the value part is C<undef> then this is converted to SQL <IS NULL>
2093 $stmt = "WHERE user = ? AND status IS NULL";
2096 To test if a column IS NOT NULL:
2100 status => { '!=', undef },
2103 =head2 Specific comparison operators
2105 If you want to specify a different type of operator for your comparison,
2106 you can use a hashref for a given column:
2110 status => { '!=', 'completed' }
2113 Which would generate:
2115 $stmt = "WHERE user = ? AND status != ?";
2116 @bind = ('nwiger', 'completed');
2118 To test against multiple values, just enclose the values in an arrayref:
2120 status => { '=', ['assigned', 'in-progress', 'pending'] };
2122 Which would give you:
2124 "WHERE status = ? OR status = ? OR status = ?"
2127 The hashref can also contain multiple pairs, in which case it is expanded
2128 into an C<AND> of its elements:
2132 status => { '!=', 'completed', -not_like => 'pending%' }
2135 # Or more dynamically, like from a form
2136 $where{user} = 'nwiger';
2137 $where{status}{'!='} = 'completed';
2138 $where{status}{'-not_like'} = 'pending%';
2140 # Both generate this
2141 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2142 @bind = ('nwiger', 'completed', 'pending%');
2145 To get an OR instead, you can combine it with the arrayref idea:
2149 priority => [ { '=', 2 }, { '>', 5 } ]
2152 Which would generate:
2154 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2155 @bind = ('2', '5', 'nwiger');
2157 If you want to include literal SQL (with or without bind values), just use a
2158 scalar reference or array reference as the value:
2161 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2162 date_expires => { '<' => \"now()" }
2165 Which would generate:
2167 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2168 @bind = ('11/26/2008');
2171 =head2 Logic and nesting operators
2173 In the example above,
2174 there is a subtle trap if you want to say something like
2175 this (notice the C<AND>):
2177 WHERE priority != ? AND priority != ?
2179 Because, in Perl you I<can't> do this:
2181 priority => { '!=', 2, '!=', 1 }
2183 As the second C<!=> key will obliterate the first. The solution
2184 is to use the special C<-modifier> form inside an arrayref:
2186 priority => [ -and => {'!=', 2},
2190 Normally, these would be joined by C<OR>, but the modifier tells it
2191 to use C<AND> instead. (Hint: You can use this in conjunction with the
2192 C<logic> option to C<new()> in order to change the way your queries
2193 work by default.) B<Important:> Note that the C<-modifier> goes
2194 B<INSIDE> the arrayref, as an extra first element. This will
2195 B<NOT> do what you think it might:
2197 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2199 Here is a quick list of equivalencies, since there is some overlap:
2202 status => {'!=', 'completed', 'not like', 'pending%' }
2203 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2206 status => {'=', ['assigned', 'in-progress']}
2207 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2208 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2212 =head2 Special operators : IN, BETWEEN, etc.
2214 You can also use the hashref format to compare a list of fields using the
2215 C<IN> comparison operator, by specifying the list as an arrayref:
2218 status => 'completed',
2219 reportid => { -in => [567, 2335, 2] }
2222 Which would generate:
2224 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2225 @bind = ('completed', '567', '2335', '2');
2227 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2230 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2231 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2232 'sqltrue' (by default : C<1=1>).
2234 In addition to the array you can supply a chunk of literal sql or
2235 literal sql with bind:
2238 customer => { -in => \[
2239 'SELECT cust_id FROM cust WHERE balance > ?',
2242 status => { -in => \'SELECT status_codes FROM states' },
2248 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2249 AND status IN ( SELECT status_codes FROM states )
2253 Finally, if the argument to C<-in> is not a reference, it will be
2254 treated as a single-element array.
2256 Another pair of operators is C<-between> and C<-not_between>,
2257 used with an arrayref of two values:
2261 completion_date => {
2262 -not_between => ['2002-10-01', '2003-02-06']
2268 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2270 Just like with C<-in> all plausible combinations of literal SQL
2274 start0 => { -between => [ 1, 2 ] },
2275 start1 => { -between => \["? AND ?", 1, 2] },
2276 start2 => { -between => \"lower(x) AND upper(y)" },
2277 start3 => { -between => [
2279 \["upper(?)", 'stuff' ],
2286 ( start0 BETWEEN ? AND ? )
2287 AND ( start1 BETWEEN ? AND ? )
2288 AND ( start2 BETWEEN lower(x) AND upper(y) )
2289 AND ( start3 BETWEEN lower(x) AND upper(?) )
2291 @bind = (1, 2, 1, 2, 'stuff');
2294 These are the two builtin "special operators"; but the
2295 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2297 =head2 Unary operators: bool
2299 If you wish to test against boolean columns or functions within your
2300 database you can use the C<-bool> and C<-not_bool> operators. For
2301 example to test the column C<is_user> being true and the column
2302 C<is_enabled> being false you would use:-
2306 -not_bool => 'is_enabled',
2311 WHERE is_user AND NOT is_enabled
2313 If a more complex combination is required, testing more conditions,
2314 then you should use the and/or operators:-
2319 -not_bool => { two=> { -rlike => 'bar' } },
2320 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2331 (NOT ( three = ? OR three > ? ))
2334 =head2 Nested conditions, -and/-or prefixes
2336 So far, we've seen how multiple conditions are joined with a top-level
2337 C<AND>. We can change this by putting the different conditions we want in
2338 hashes and then putting those hashes in an array. For example:
2343 status => { -like => ['pending%', 'dispatched'] },
2347 status => 'unassigned',
2351 This data structure would create the following:
2353 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2354 OR ( user = ? AND status = ? ) )";
2355 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2358 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2359 to change the logic inside :
2365 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2366 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2373 WHERE ( user = ? AND (
2374 ( workhrs > ? AND geo = ? )
2375 OR ( workhrs < ? OR geo = ? )
2378 =head3 Algebraic inconsistency, for historical reasons
2380 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2381 operator goes C<outside> of the nested structure; whereas when connecting
2382 several constraints on one column, the C<-and> operator goes
2383 C<inside> the arrayref. Here is an example combining both features :
2386 -and => [a => 1, b => 2],
2387 -or => [c => 3, d => 4],
2388 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2393 WHERE ( ( ( a = ? AND b = ? )
2394 OR ( c = ? OR d = ? )
2395 OR ( e LIKE ? AND e LIKE ? ) ) )
2397 This difference in syntax is unfortunate but must be preserved for
2398 historical reasons. So be careful : the two examples below would
2399 seem algebraically equivalent, but they are not
2401 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2402 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2404 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2405 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2408 =head2 Literal SQL and value type operators
2410 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2411 side" is a column name and the "right side" is a value (normally rendered as
2412 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2413 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2414 alter this behavior. There are several ways of doing so.
2418 This is a virtual operator that signals the string to its right side is an
2419 identifier (a column name) and not a value. For example to compare two
2420 columns you would write:
2423 priority => { '<', 2 },
2424 requestor => { -ident => 'submitter' },
2429 $stmt = "WHERE priority < ? AND requestor = submitter";
2432 If you are maintaining legacy code you may see a different construct as
2433 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2438 This is a virtual operator that signals that the construct to its right side
2439 is a value to be passed to DBI. This is for example necessary when you want
2440 to write a where clause against an array (for RDBMS that support such
2441 datatypes). For example:
2444 array => { -value => [1, 2, 3] }
2449 $stmt = 'WHERE array = ?';
2450 @bind = ([1, 2, 3]);
2452 Note that if you were to simply say:
2458 the result would probably not be what you wanted:
2460 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2465 Finally, sometimes only literal SQL will do. To include a random snippet
2466 of SQL verbatim, you specify it as a scalar reference. Consider this only
2467 as a last resort. Usually there is a better way. For example:
2470 priority => { '<', 2 },
2471 requestor => { -in => \'(SELECT name FROM hitmen)' },
2476 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2479 Note that in this example, you only get one bind parameter back, since
2480 the verbatim SQL is passed as part of the statement.
2484 Never use untrusted input as a literal SQL argument - this is a massive
2485 security risk (there is no way to check literal snippets for SQL
2486 injections and other nastyness). If you need to deal with untrusted input
2487 use literal SQL with placeholders as described next.
2489 =head3 Literal SQL with placeholders and bind values (subqueries)
2491 If the literal SQL to be inserted has placeholders and bind values,
2492 use a reference to an arrayref (yes this is a double reference --
2493 not so common, but perfectly legal Perl). For example, to find a date
2494 in Postgres you can use something like this:
2497 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2502 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2505 Note that you must pass the bind values in the same format as they are returned
2506 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2507 provide the bind values in the C<< [ column_meta => value ] >> format, where
2508 C<column_meta> is an opaque scalar value; most commonly the column name, but
2509 you can use any scalar value (including references and blessed references),
2510 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2511 to C<columns> the above example will look like:
2514 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2517 Literal SQL is especially useful for nesting parenthesized clauses in the
2518 main SQL query. Here is a first example :
2520 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2524 bar => \["IN ($sub_stmt)" => @sub_bind],
2529 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2530 WHERE c2 < ? AND c3 LIKE ?))";
2531 @bind = (1234, 100, "foo%");
2533 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2534 are expressed in the same way. Of course the C<$sub_stmt> and
2535 its associated bind values can be generated through a former call
2538 my ($sub_stmt, @sub_bind)
2539 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2540 c3 => {-like => "foo%"}});
2543 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2546 In the examples above, the subquery was used as an operator on a column;
2547 but the same principle also applies for a clause within the main C<%where>
2548 hash, like an EXISTS subquery :
2550 my ($sub_stmt, @sub_bind)
2551 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2552 my %where = ( -and => [
2554 \["EXISTS ($sub_stmt)" => @sub_bind],
2559 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2560 WHERE c1 = ? AND c2 > t0.c0))";
2564 Observe that the condition on C<c2> in the subquery refers to
2565 column C<t0.c0> of the main query : this is I<not> a bind
2566 value, so we have to express it through a scalar ref.
2567 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2568 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2569 what we wanted here.
2571 Finally, here is an example where a subquery is used
2572 for expressing unary negation:
2574 my ($sub_stmt, @sub_bind)
2575 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2576 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2578 lname => {like => '%son%'},
2579 \["NOT ($sub_stmt)" => @sub_bind],
2584 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2585 @bind = ('%son%', 10, 20)
2587 =head3 Deprecated usage of Literal SQL
2589 Below are some examples of archaic use of literal SQL. It is shown only as
2590 reference for those who deal with legacy code. Each example has a much
2591 better, cleaner and safer alternative that users should opt for in new code.
2597 my %where = ( requestor => \'IS NOT NULL' )
2599 $stmt = "WHERE requestor IS NOT NULL"
2601 This used to be the way of generating NULL comparisons, before the handling
2602 of C<undef> got formalized. For new code please use the superior syntax as
2603 described in L</Tests for NULL values>.
2607 my %where = ( requestor => \'= submitter' )
2609 $stmt = "WHERE requestor = submitter"
2611 This used to be the only way to compare columns. Use the superior L</-ident>
2612 method for all new code. For example an identifier declared in such a way
2613 will be properly quoted if L</quote_char> is properly set, while the legacy
2614 form will remain as supplied.
2618 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2620 $stmt = "WHERE completed > ? AND is_ready"
2621 @bind = ('2012-12-21')
2623 Using an empty string literal used to be the only way to express a boolean.
2624 For all new code please use the much more readable
2625 L<-bool|/Unary operators: bool> operator.
2631 These pages could go on for a while, since the nesting of the data
2632 structures this module can handle are pretty much unlimited (the
2633 module implements the C<WHERE> expansion as a recursive function
2634 internally). Your best bet is to "play around" with the module a
2635 little to see how the data structures behave, and choose the best
2636 format for your data based on that.
2638 And of course, all the values above will probably be replaced with
2639 variables gotten from forms or the command line. After all, if you
2640 knew everything ahead of time, you wouldn't have to worry about
2641 dynamically-generating SQL and could just hardwire it into your
2644 =head1 ORDER BY CLAUSES
2646 Some functions take an order by clause. This can either be a scalar (just a
2647 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2648 or an array of either of the two previous forms. Examples:
2650 Given | Will Generate
2651 ----------------------------------------------------------
2653 \'colA DESC' | ORDER BY colA DESC
2655 'colA' | ORDER BY colA
2657 [qw/colA colB/] | ORDER BY colA, colB
2659 {-asc => 'colA'} | ORDER BY colA ASC
2661 {-desc => 'colB'} | ORDER BY colB DESC
2663 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2665 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2668 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2669 { -desc => [qw/colB/], | colC ASC, colD ASC
2670 { -asc => [qw/colC colD/],|
2672 ===========================================================
2676 =head1 SPECIAL OPERATORS
2678 my $sqlmaker = SQL::Abstract->new(special_ops => [
2682 my ($self, $field, $op, $arg) = @_;
2688 handler => 'method_name',
2692 A "special operator" is a SQL syntactic clause that can be
2693 applied to a field, instead of a usual binary operator.
2696 WHERE field IN (?, ?, ?)
2697 WHERE field BETWEEN ? AND ?
2698 WHERE MATCH(field) AGAINST (?, ?)
2700 Special operators IN and BETWEEN are fairly standard and therefore
2701 are builtin within C<SQL::Abstract> (as the overridable methods
2702 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2703 like the MATCH .. AGAINST example above which is specific to MySQL,
2704 you can write your own operator handlers - supply a C<special_ops>
2705 argument to the C<new> method. That argument takes an arrayref of
2706 operator definitions; each operator definition is a hashref with two
2713 the regular expression to match the operator
2717 Either a coderef or a plain scalar method name. In both cases
2718 the expected return is C<< ($sql, @bind) >>.
2720 When supplied with a method name, it is simply called on the
2721 L<SQL::Abstract/> object as:
2723 $self->$method_name ($field, $op, $arg)
2727 $op is the part that matched the handler regex
2728 $field is the LHS of the operator
2731 When supplied with a coderef, it is called as:
2733 $coderef->($self, $field, $op, $arg)
2738 For example, here is an implementation
2739 of the MATCH .. AGAINST syntax for MySQL
2741 my $sqlmaker = SQL::Abstract->new(special_ops => [
2743 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2744 {regex => qr/^match$/i,
2746 my ($self, $field, $op, $arg) = @_;
2747 $arg = [$arg] if not ref $arg;
2748 my $label = $self->_quote($field);
2749 my ($placeholder) = $self->_convert('?');
2750 my $placeholders = join ", ", (($placeholder) x @$arg);
2751 my $sql = $self->_sqlcase('match') . " ($label) "
2752 . $self->_sqlcase('against') . " ($placeholders) ";
2753 my @bind = $self->_bindtype($field, @$arg);
2754 return ($sql, @bind);
2761 =head1 UNARY OPERATORS
2763 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2767 my ($self, $op, $arg) = @_;
2773 handler => 'method_name',
2777 A "unary operator" is a SQL syntactic clause that can be
2778 applied to a field - the operator goes before the field
2780 You can write your own operator handlers - supply a C<unary_ops>
2781 argument to the C<new> method. That argument takes an arrayref of
2782 operator definitions; each operator definition is a hashref with two
2789 the regular expression to match the operator
2793 Either a coderef or a plain scalar method name. In both cases
2794 the expected return is C<< $sql >>.
2796 When supplied with a method name, it is simply called on the
2797 L<SQL::Abstract/> object as:
2799 $self->$method_name ($op, $arg)
2803 $op is the part that matched the handler regex
2804 $arg is the RHS or argument of the operator
2806 When supplied with a coderef, it is called as:
2808 $coderef->($self, $op, $arg)
2816 Thanks to some benchmarking by Mark Stosberg, it turns out that
2817 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2818 I must admit this wasn't an intentional design issue, but it's a
2819 byproduct of the fact that you get to control your C<DBI> handles
2822 To maximize performance, use a code snippet like the following:
2824 # prepare a statement handle using the first row
2825 # and then reuse it for the rest of the rows
2827 for my $href (@array_of_hashrefs) {
2828 $stmt ||= $sql->insert('table', $href);
2829 $sth ||= $dbh->prepare($stmt);
2830 $sth->execute($sql->values($href));
2833 The reason this works is because the keys in your C<$href> are sorted
2834 internally by B<SQL::Abstract>. Thus, as long as your data retains
2835 the same structure, you only have to generate the SQL the first time
2836 around. On subsequent queries, simply use the C<values> function provided
2837 by this module to return your values in the correct order.
2839 However this depends on the values having the same type - if, for
2840 example, the values of a where clause may either have values
2841 (resulting in sql of the form C<column = ?> with a single bind
2842 value), or alternatively the values might be C<undef> (resulting in
2843 sql of the form C<column IS NULL> with no bind value) then the
2844 caching technique suggested will not work.
2848 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2849 really like this part (I do, at least). Building up a complex query
2850 can be as simple as the following:
2857 use CGI::FormBuilder;
2860 my $form = CGI::FormBuilder->new(...);
2861 my $sql = SQL::Abstract->new;
2863 if ($form->submitted) {
2864 my $field = $form->field;
2865 my $id = delete $field->{id};
2866 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2869 Of course, you would still have to connect using C<DBI> to run the
2870 query, but the point is that if you make your form look like your
2871 table, the actual query script can be extremely simplistic.
2873 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2874 a fast interface to returning and formatting data. I frequently
2875 use these three modules together to write complex database query
2876 apps in under 50 lines.
2882 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2884 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2890 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2891 Great care has been taken to preserve the I<published> behavior
2892 documented in previous versions in the 1.* family; however,
2893 some features that were previously undocumented, or behaved
2894 differently from the documentation, had to be changed in order
2895 to clarify the semantics. Hence, client code that was relying
2896 on some dark areas of C<SQL::Abstract> v1.*
2897 B<might behave differently> in v1.50.
2899 The main changes are :
2905 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2909 support for the { operator => \"..." } construct (to embed literal SQL)
2913 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2917 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2921 defensive programming : check arguments
2925 fixed bug with global logic, which was previously implemented
2926 through global variables yielding side-effects. Prior versions would
2927 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2928 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2929 Now this is interpreted
2930 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2935 fixed semantics of _bindtype on array args
2939 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2940 we just avoid shifting arrays within that tree.
2944 dropped the C<_modlogic> function
2948 =head1 ACKNOWLEDGEMENTS
2950 There are a number of individuals that have really helped out with
2951 this module. Unfortunately, most of them submitted bugs via CPAN
2952 so I have no idea who they are! But the people I do know are:
2954 Ash Berlin (order_by hash term support)
2955 Matt Trout (DBIx::Class support)
2956 Mark Stosberg (benchmarking)
2957 Chas Owens (initial "IN" operator support)
2958 Philip Collins (per-field SQL functions)
2959 Eric Kolve (hashref "AND" support)
2960 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2961 Dan Kubb (support for "quote_char" and "name_sep")
2962 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2963 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2964 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2965 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2966 Oliver Charles (support for "RETURNING" after "INSERT")
2972 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2976 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2978 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2980 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2981 While not an official support venue, C<DBIx::Class> makes heavy use of
2982 C<SQL::Abstract>, and as such list members there are very familiar with
2983 how to create queries.
2987 This module is free software; you may copy this under the same
2988 terms as perl itself (either the GNU General Public License or
2989 the Artistic License)