1 package SQL::Abstract; # see doc at end of file
3 # LDNOTE : this code is heavy refactoring from original SQLA.
4 # Several design decisions will need discussion during
5 # the test / diffusion / acceptance phase; those are marked with flag
6 # 'LDNOTE' (note by laurent.dami AT free.fr)
14 #======================================================================
16 #======================================================================
18 our $VERSION = '1.73';
20 # This would confuse some packagers
21 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
25 # special operators (-in, -between). May be extended/overridden by user.
26 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
27 my @BUILTIN_SPECIAL_OPS = (
28 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
29 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
30 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
31 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
34 # unaryish operators - key maps to handler
35 my @BUILTIN_UNARY_OPS = (
36 # the digits are backcompat stuff
37 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
38 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
39 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
40 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
41 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
42 { regex => qr/^ value $/ix, handler => '_where_op_VALUE' },
45 #======================================================================
46 # DEBUGGING AND ERROR REPORTING
47 #======================================================================
50 return unless $_[0]->{debug}; shift; # a little faster
51 my $func = (caller(1))[3];
52 warn "[$func] ", @_, "\n";
56 my($func) = (caller(1))[3];
57 Carp::carp "[$func] Warning: ", @_;
61 my($func) = (caller(1))[3];
62 Carp::croak "[$func] Fatal: ", @_;
66 #======================================================================
68 #======================================================================
72 my $class = ref($self) || $self;
73 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
75 # choose our case by keeping an option around
76 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
78 # default logic for interpreting arrayrefs
79 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
81 # how to return bind vars
82 # LDNOTE: changed nwiger code : why this 'delete' ??
83 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
84 $opt{bindtype} ||= 'normal';
86 # default comparison is "=", but can be overridden
89 # try to recognize which are the 'equality' and 'unequality' ops
90 # (temporary quickfix, should go through a more seasoned API)
91 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
92 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
95 $opt{sqltrue} ||= '1=1';
96 $opt{sqlfalse} ||= '0=1';
99 $opt{special_ops} ||= [];
100 # regexes are applied in order, thus push after user-defines
101 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
104 $opt{unary_ops} ||= [];
105 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
107 # rudimentary saniy-check for user supplied bits treated as functions/operators
108 # If a purported function matches this regular expression, an exception is thrown.
109 # Literal SQL is *NOT* subject to this check, only functions (and column names
110 # when quoting is not in effect)
113 # need to guard against ()'s in column names too, but this will break tons of
114 # hacks... ideas anyone?
115 $opt{injection_guard} ||= qr/
121 return bless \%opt, $class;
125 sub _assert_pass_injection_guard {
126 if ($_[1] =~ $_[0]->{injection_guard}) {
127 my $class = ref $_[0];
128 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
129 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
130 . "{injection_guard} attribute to ${class}->new()"
135 #======================================================================
137 #======================================================================
141 my $table = $self->_table(shift);
142 my $data = shift || return;
145 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
146 my ($sql, @bind) = $self->$method($data);
147 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
149 if ($options->{returning}) {
150 my ($s, @b) = $self->_insert_returning ($options);
155 return wantarray ? ($sql, @bind) : $sql;
158 sub _insert_returning {
159 my ($self, $options) = @_;
161 my $f = $options->{returning};
163 my $fieldlist = $self->_SWITCH_refkind($f, {
164 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
165 SCALAR => sub {$self->_quote($f)},
166 SCALARREF => sub {$$f},
168 return $self->_sqlcase(' returning ') . $fieldlist;
171 sub _insert_HASHREF { # explicit list of fields and then values
172 my ($self, $data) = @_;
174 my @fields = sort keys %$data;
176 my ($sql, @bind) = $self->_insert_values($data);
179 $_ = $self->_quote($_) foreach @fields;
180 $sql = "( ".join(", ", @fields).") ".$sql;
182 return ($sql, @bind);
185 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
186 my ($self, $data) = @_;
188 # no names (arrayref) so can't generate bindtype
189 $self->{bindtype} ne 'columns'
190 or belch "can't do 'columns' bindtype when called with arrayref";
192 # fold the list of values into a hash of column name - value pairs
193 # (where the column names are artificially generated, and their
194 # lexicographical ordering keep the ordering of the original list)
195 my $i = "a"; # incremented values will be in lexicographical order
196 my $data_in_hash = { map { ($i++ => $_) } @$data };
198 return $self->_insert_values($data_in_hash);
201 sub _insert_ARRAYREFREF { # literal SQL with bind
202 my ($self, $data) = @_;
204 my ($sql, @bind) = @${$data};
205 $self->_assert_bindval_matches_bindtype(@bind);
207 return ($sql, @bind);
211 sub _insert_SCALARREF { # literal SQL without bind
212 my ($self, $data) = @_;
218 my ($self, $data) = @_;
220 my (@values, @all_bind);
221 foreach my $column (sort keys %$data) {
222 my $v = $data->{$column};
224 $self->_SWITCH_refkind($v, {
227 if ($self->{array_datatypes}) { # if array datatype are activated
229 push @all_bind, $self->_bindtype($column, $v);
231 else { # else literal SQL with bind
232 my ($sql, @bind) = @$v;
233 $self->_assert_bindval_matches_bindtype(@bind);
235 push @all_bind, @bind;
239 ARRAYREFREF => sub { # literal SQL with bind
240 my ($sql, @bind) = @${$v};
241 $self->_assert_bindval_matches_bindtype(@bind);
243 push @all_bind, @bind;
246 # THINK : anything useful to do with a HASHREF ?
247 HASHREF => sub { # (nothing, but old SQLA passed it through)
248 #TODO in SQLA >= 2.0 it will die instead
249 belch "HASH ref as bind value in insert is not supported";
251 push @all_bind, $self->_bindtype($column, $v);
254 SCALARREF => sub { # literal SQL without bind
258 SCALAR_or_UNDEF => sub {
260 push @all_bind, $self->_bindtype($column, $v);
267 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
268 return ($sql, @all_bind);
273 #======================================================================
275 #======================================================================
280 my $table = $self->_table(shift);
281 my $data = shift || return;
284 # first build the 'SET' part of the sql statement
285 my (@set, @all_bind);
286 puke "Unsupported data type specified to \$sql->update"
287 unless ref $data eq 'HASH';
289 for my $k (sort keys %$data) {
292 my $label = $self->_quote($k);
294 $self->_SWITCH_refkind($v, {
296 if ($self->{array_datatypes}) { # array datatype
297 push @set, "$label = ?";
298 push @all_bind, $self->_bindtype($k, $v);
300 else { # literal SQL with bind
301 my ($sql, @bind) = @$v;
302 $self->_assert_bindval_matches_bindtype(@bind);
303 push @set, "$label = $sql";
304 push @all_bind, @bind;
307 ARRAYREFREF => sub { # literal SQL with bind
308 my ($sql, @bind) = @${$v};
309 $self->_assert_bindval_matches_bindtype(@bind);
310 push @set, "$label = $sql";
311 push @all_bind, @bind;
313 SCALARREF => sub { # literal SQL without bind
314 push @set, "$label = $$v";
317 my ($op, $arg, @rest) = %$v;
319 puke 'Operator calls in update must be in the form { -op => $arg }'
320 if (@rest or not $op =~ /^\-(.+)/);
322 local $self->{_nested_func_lhs} = $k;
323 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
325 push @set, "$label = $sql";
326 push @all_bind, @bind;
328 SCALAR_or_UNDEF => sub {
329 push @set, "$label = ?";
330 push @all_bind, $self->_bindtype($k, $v);
336 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
340 my($where_sql, @where_bind) = $self->where($where);
342 push @all_bind, @where_bind;
345 return wantarray ? ($sql, @all_bind) : $sql;
351 #======================================================================
353 #======================================================================
358 my $table = $self->_table(shift);
359 my $fields = shift || '*';
363 my($where_sql, @bind) = $self->where($where, $order);
365 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
367 my $sql = join(' ', $self->_sqlcase('select'), $f,
368 $self->_sqlcase('from'), $table)
371 return wantarray ? ($sql, @bind) : $sql;
374 #======================================================================
376 #======================================================================
381 my $table = $self->_table(shift);
385 my($where_sql, @bind) = $self->where($where);
386 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
388 return wantarray ? ($sql, @bind) : $sql;
392 #======================================================================
394 #======================================================================
398 # Finally, a separate routine just to handle WHERE clauses
400 my ($self, $where, $order) = @_;
403 my ($sql, @bind) = $self->_recurse_where($where);
404 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
408 $sql .= $self->_order_by($order);
411 return wantarray ? ($sql, @bind) : $sql;
416 my ($self, $where, $logic) = @_;
418 # dispatch on appropriate method according to refkind of $where
419 my $method = $self->_METHOD_FOR_refkind("_where", $where);
421 my ($sql, @bind) = $self->$method($where, $logic);
423 # DBIx::Class directly calls _recurse_where in scalar context, so
424 # we must implement it, even if not in the official API
425 return wantarray ? ($sql, @bind) : $sql;
430 #======================================================================
431 # WHERE: top-level ARRAYREF
432 #======================================================================
435 sub _where_ARRAYREF {
436 my ($self, $where, $logic) = @_;
438 $logic = uc($logic || $self->{logic});
439 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
441 my @clauses = @$where;
443 my (@sql_clauses, @all_bind);
444 # need to use while() so can shift() for pairs
445 while (my $el = shift @clauses) {
447 # switch according to kind of $el and get corresponding ($sql, @bind)
448 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
450 # skip empty elements, otherwise get invalid trailing AND stuff
451 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
455 $self->_assert_bindval_matches_bindtype(@b);
459 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
460 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
461 # side-effect: the first hashref within an array would change
462 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
463 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
464 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
466 SCALARREF => sub { ($$el); },
468 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
469 $self->_recurse_where({$el => shift(@clauses)})},
471 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
475 push @sql_clauses, $sql;
476 push @all_bind, @bind;
480 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
483 #======================================================================
484 # WHERE: top-level ARRAYREFREF
485 #======================================================================
487 sub _where_ARRAYREFREF {
488 my ($self, $where) = @_;
489 my ($sql, @bind) = @$$where;
490 $self->_assert_bindval_matches_bindtype(@bind);
491 return ($sql, @bind);
494 #======================================================================
495 # WHERE: top-level HASHREF
496 #======================================================================
499 my ($self, $where) = @_;
500 my (@sql_clauses, @all_bind);
502 for my $k (sort keys %$where) {
503 my $v = $where->{$k};
505 # ($k => $v) is either a special unary op or a regular hashpair
506 my ($sql, @bind) = do {
508 # put the operator in canonical form
510 $op = substr $op, 1; # remove initial dash
511 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
512 $op =~ s/\s+/ /g; # compress whitespace
514 # so that -not_foo works correctly
515 $op =~ s/^not_/NOT /i;
517 $self->_debug("Unary OP(-$op) within hashref, recursing...");
518 my ($s, @b) = $self->_where_unary_op ($op, $v);
520 # top level vs nested
521 # we assume that handled unary ops will take care of their ()s
523 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
525 defined($self->{_nested_func_lhs}) && ($self->{_nested_func_lhs} eq $k)
530 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
531 $self->$method($k, $v);
535 push @sql_clauses, $sql;
536 push @all_bind, @bind;
539 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
542 sub _where_unary_op {
543 my ($self, $op, $rhs) = @_;
545 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
546 my $handler = $op_entry->{handler};
548 if (not ref $handler) {
549 if ($op =~ s/ [_\s]? \d+ $//x ) {
550 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
551 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
553 return $self->$handler ($op, $rhs);
555 elsif (ref $handler eq 'CODE') {
556 return $handler->($self, $op, $rhs);
559 puke "Illegal handler for operator $op - expecting a method name or a coderef";
563 $self->_debug("Generic unary OP: $op - recursing as function");
565 $self->_assert_pass_injection_guard($op);
567 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
569 puke "Illegal use of top-level '$op'"
570 unless $self->{_nested_func_lhs};
573 $self->_convert('?'),
574 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
578 $self->_recurse_where ($rhs)
582 $sql = sprintf ('%s %s',
583 $self->_sqlcase($op),
587 return ($sql, @bind);
590 sub _where_op_ANDOR {
591 my ($self, $op, $v) = @_;
593 $self->_SWITCH_refkind($v, {
595 return $self->_where_ARRAYREF($v, $op);
599 return ( $op =~ /^or/i )
600 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
601 : $self->_where_HASHREF($v);
605 puke "-$op => \\\$scalar makes little sense, use " .
607 ? '[ \$scalar, \%rest_of_conditions ] instead'
608 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
613 puke "-$op => \\[...] makes little sense, use " .
615 ? '[ \[...], \%rest_of_conditions ] instead'
616 : '-and => [ \[...], \%rest_of_conditions ] instead'
620 SCALAR => sub { # permissively interpreted as SQL
621 puke "-$op => \$value makes little sense, use -bool => \$value instead";
625 puke "-$op => undef not supported";
631 my ($self, $op, $v) = @_;
633 $self->_SWITCH_refkind($v, {
635 SCALAR => sub { # permissively interpreted as SQL
636 belch "literal SQL should be -nest => \\'scalar' "
637 . "instead of -nest => 'scalar' ";
642 puke "-$op => undef not supported";
646 $self->_recurse_where ($v);
654 my ($self, $op, $v) = @_;
656 my ($s, @b) = $self->_SWITCH_refkind($v, {
657 SCALAR => sub { # interpreted as SQL column
658 $self->_convert($self->_quote($v));
662 puke "-$op => undef not supported";
666 $self->_recurse_where ($v);
670 $s = "(NOT $s)" if $op =~ /^not/i;
675 sub _where_op_IDENT {
677 my ($op, $rhs) = splice @_, -2;
679 puke "-$op takes a single scalar argument (a quotable identifier)";
682 # in case we are called as a top level special op (no '=')
685 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
693 sub _where_op_VALUE {
695 my ($op, $rhs) = splice @_, -2;
697 # in case we are called as a top level special op (no '=')
702 ($lhs || $self->{_nested_func_lhs}),
709 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
713 $self->_convert('?'),
719 sub _where_hashpair_ARRAYREF {
720 my ($self, $k, $v) = @_;
723 my @v = @$v; # need copy because of shift below
724 $self->_debug("ARRAY($k) means distribute over elements");
726 # put apart first element if it is an operator (-and, -or)
728 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
732 my @distributed = map { {$k => $_} } @v;
735 $self->_debug("OP($op) reinjected into the distributed array");
736 unshift @distributed, $op;
739 my $logic = $op ? substr($op, 1) : '';
741 return $self->_recurse_where(\@distributed, $logic);
744 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
745 $self->_debug("empty ARRAY($k) means 0=1");
746 return ($self->{sqlfalse});
750 sub _where_hashpair_HASHREF {
751 my ($self, $k, $v, $logic) = @_;
754 local $self->{_nested_func_lhs} = $self->{_nested_func_lhs};
756 my ($all_sql, @all_bind);
758 for my $orig_op (sort keys %$v) {
759 my $val = $v->{$orig_op};
761 # put the operator in canonical form
764 # FIXME - we need to phase out dash-less ops
765 $op =~ s/^-//; # remove possible initial dash
766 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
767 $op =~ s/\s+/ /g; # compress whitespace
769 $self->_assert_pass_injection_guard($op);
771 # so that -not_foo works correctly
772 $op =~ s/^not_/NOT /i;
776 # CASE: col-value logic modifiers
777 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
778 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
780 # CASE: special operators like -in or -between
781 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
782 my $handler = $special_op->{handler};
784 puke "No handler supplied for special operator $orig_op";
786 elsif (not ref $handler) {
787 ($sql, @bind) = $self->$handler ($k, $op, $val);
789 elsif (ref $handler eq 'CODE') {
790 ($sql, @bind) = $handler->($self, $k, $op, $val);
793 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
797 $self->_SWITCH_refkind($val, {
799 ARRAYREF => sub { # CASE: col => {op => \@vals}
800 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
803 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
804 my ($sub_sql, @sub_bind) = @$$val;
805 $self->_assert_bindval_matches_bindtype(@sub_bind);
806 $sql = join ' ', $self->_convert($self->_quote($k)),
807 $self->_sqlcase($op),
812 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
813 my $is = ($op =~ $self->{equality_op}) ? 'is' :
814 ($op =~ $self->{inequality_op}) ? 'is not' :
815 puke "unexpected operator '$orig_op' with undef operand";
816 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
819 FALLBACK => sub { # CASE: col => {op/func => $stuff}
821 # retain for proper column type bind
822 $self->{_nested_func_lhs} ||= $k;
824 ($sql, @bind) = $self->_where_unary_op ($op, $val);
827 $self->_convert($self->_quote($k)),
828 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
834 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
835 push @all_bind, @bind;
837 return ($all_sql, @all_bind);
842 sub _where_field_op_ARRAYREF {
843 my ($self, $k, $op, $vals) = @_;
845 my @vals = @$vals; #always work on a copy
848 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
850 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
853 # see if the first element is an -and/-or op
855 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
860 # distribute $op over each remaining member of @vals, append logic if exists
861 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
863 # LDNOTE : had planned to change the distribution logic when
864 # $op =~ $self->{inequality_op}, because of Morgan laws :
865 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
866 # WHERE field != 22 OR field != 33 : the user probably means
867 # WHERE field != 22 AND field != 33.
868 # To do this, replace the above to roughly :
869 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
870 # return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
874 # try to DWIM on equality operators
875 # LDNOTE : not 100% sure this is the correct thing to do ...
876 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
877 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
880 puke "operator '$op' applied on an empty array (field '$k')";
885 sub _where_hashpair_SCALARREF {
886 my ($self, $k, $v) = @_;
887 $self->_debug("SCALAR($k) means literal SQL: $$v");
888 my $sql = $self->_quote($k) . " " . $$v;
892 # literal SQL with bind
893 sub _where_hashpair_ARRAYREFREF {
894 my ($self, $k, $v) = @_;
895 $self->_debug("REF($k) means literal SQL: @${$v}");
896 my ($sql, @bind) = @$$v;
897 $self->_assert_bindval_matches_bindtype(@bind);
898 $sql = $self->_quote($k) . " " . $sql;
899 return ($sql, @bind );
902 # literal SQL without bind
903 sub _where_hashpair_SCALAR {
904 my ($self, $k, $v) = @_;
905 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
906 my $sql = join ' ', $self->_convert($self->_quote($k)),
907 $self->_sqlcase($self->{cmp}),
908 $self->_convert('?');
909 my @bind = $self->_bindtype($k, $v);
910 return ( $sql, @bind);
914 sub _where_hashpair_UNDEF {
915 my ($self, $k, $v) = @_;
916 $self->_debug("UNDEF($k) means IS NULL");
917 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
921 #======================================================================
922 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
923 #======================================================================
926 sub _where_SCALARREF {
927 my ($self, $where) = @_;
930 $self->_debug("SCALAR(*top) means literal SQL: $$where");
936 my ($self, $where) = @_;
939 $self->_debug("NOREF(*top) means literal SQL: $where");
950 #======================================================================
951 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
952 #======================================================================
955 sub _where_field_BETWEEN {
956 my ($self, $k, $op, $vals) = @_;
958 my ($label, $and, $placeholder);
959 $label = $self->_convert($self->_quote($k));
960 $and = ' ' . $self->_sqlcase('and') . ' ';
961 $placeholder = $self->_convert('?');
962 $op = $self->_sqlcase($op);
964 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
966 my ($s, @b) = @$$vals;
967 $self->_assert_bindval_matches_bindtype(@b);
974 puke "special op 'between' accepts an arrayref with exactly two values"
977 my (@all_sql, @all_bind);
978 foreach my $val (@$vals) {
979 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
981 return ($placeholder, $self->_bindtype($k, $val) );
987 my ($sql, @bind) = @$$val;
988 $self->_assert_bindval_matches_bindtype(@bind);
989 return ($sql, @bind);
992 my ($func, $arg, @rest) = %$val;
993 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
994 if (@rest or $func !~ /^ \- (.+)/x);
995 local $self->{_nested_func_lhs} = $k;
996 $self->_where_unary_op ($1 => $arg);
1000 push @all_bind, @bind;
1004 (join $and, @all_sql),
1009 puke "special op 'between' accepts an arrayref with two values, or a single literal scalarref/arrayref-ref";
1013 my $sql = "( $label $op $clause )";
1014 return ($sql, @bind)
1018 sub _where_field_IN {
1019 my ($self, $k, $op, $vals) = @_;
1021 # backwards compatibility : if scalar, force into an arrayref
1022 $vals = [$vals] if defined $vals && ! ref $vals;
1024 my ($label) = $self->_convert($self->_quote($k));
1025 my ($placeholder) = $self->_convert('?');
1026 $op = $self->_sqlcase($op);
1028 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1029 ARRAYREF => sub { # list of choices
1030 if (@$vals) { # nonempty list
1031 my (@all_sql, @all_bind);
1033 for my $val (@$vals) {
1034 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1036 return ($placeholder, $val);
1041 ARRAYREFREF => sub {
1042 my ($sql, @bind) = @$$val;
1043 $self->_assert_bindval_matches_bindtype(@bind);
1044 return ($sql, @bind);
1047 my ($func, $arg, @rest) = %$val;
1048 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1049 if (@rest or $func !~ /^ \- (.+)/x);
1050 local $self->{_nested_func_lhs} = $k;
1051 $self->_where_unary_op ($1 => $arg);
1054 return $self->_sqlcase('null');
1057 push @all_sql, $sql;
1058 push @all_bind, @bind;
1062 sprintf ('%s %s ( %s )',
1065 join (', ', @all_sql)
1067 $self->_bindtype($k, @all_bind),
1070 else { # empty list : some databases won't understand "IN ()", so DWIM
1071 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1076 SCALARREF => sub { # literal SQL
1077 my $sql = $self->_open_outer_paren ($$vals);
1078 return ("$label $op ( $sql )");
1080 ARRAYREFREF => sub { # literal SQL with bind
1081 my ($sql, @bind) = @$$vals;
1082 $self->_assert_bindval_matches_bindtype(@bind);
1083 $sql = $self->_open_outer_paren ($sql);
1084 return ("$label $op ( $sql )", @bind);
1088 puke "special op 'in' requires an arrayref (or scalarref/arrayref-ref)";
1092 return ($sql, @bind);
1095 # Some databases (SQLite) treat col IN (1, 2) different from
1096 # col IN ( (1, 2) ). Use this to strip all outer parens while
1097 # adding them back in the corresponding method
1098 sub _open_outer_paren {
1099 my ($self, $sql) = @_;
1100 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/xs;
1105 #======================================================================
1107 #======================================================================
1110 my ($self, $arg) = @_;
1113 for my $c ($self->_order_by_chunks ($arg) ) {
1114 $self->_SWITCH_refkind ($c, {
1115 SCALAR => sub { push @sql, $c },
1116 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1122 $self->_sqlcase(' order by'),
1128 return wantarray ? ($sql, @bind) : $sql;
1131 sub _order_by_chunks {
1132 my ($self, $arg) = @_;
1134 return $self->_SWITCH_refkind($arg, {
1137 map { $self->_order_by_chunks ($_ ) } @$arg;
1140 ARRAYREFREF => sub {
1141 my ($s, @b) = @$$arg;
1142 $self->_assert_bindval_matches_bindtype(@b);
1146 SCALAR => sub {$self->_quote($arg)},
1148 UNDEF => sub {return () },
1150 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1153 # get first pair in hash
1154 my ($key, $val, @rest) = %$arg;
1156 return () unless $key;
1158 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1159 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1165 for my $c ($self->_order_by_chunks ($val)) {
1168 $self->_SWITCH_refkind ($c, {
1173 ($sql, @bind) = @$c;
1177 $sql = $sql . ' ' . $self->_sqlcase($direction);
1179 push @ret, [ $sql, @bind];
1188 #======================================================================
1189 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1190 #======================================================================
1195 $self->_SWITCH_refkind($from, {
1196 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1197 SCALAR => sub {$self->_quote($from)},
1198 SCALARREF => sub {$$from},
1203 #======================================================================
1205 #======================================================================
1207 # highly optimized, as it's called way too often
1209 # my ($self, $label) = @_;
1211 return '' unless defined $_[1];
1212 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1214 unless ($_[0]->{quote_char}) {
1215 $_[0]->_assert_pass_injection_guard($_[1]);
1219 my $qref = ref $_[0]->{quote_char};
1222 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1224 elsif ($qref eq 'ARRAY') {
1225 ($l, $r) = @{$_[0]->{quote_char}};
1228 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1231 # parts containing * are naturally unquoted
1232 return join( $_[0]->{name_sep}||'', map
1233 { $_ eq '*' ? $_ : $l . $_ . $r }
1234 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1239 # Conversion, if applicable
1241 #my ($self, $arg) = @_;
1243 # LDNOTE : modified the previous implementation below because
1244 # it was not consistent : the first "return" is always an array,
1245 # the second "return" is context-dependent. Anyway, _convert
1246 # seems always used with just a single argument, so make it a
1248 # return @_ unless $self->{convert};
1249 # my $conv = $self->_sqlcase($self->{convert});
1250 # my @ret = map { $conv.'('.$_.')' } @_;
1251 # return wantarray ? @ret : $ret[0];
1252 if ($_[0]->{convert}) {
1253 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1260 #my ($self, $col, @vals) = @_;
1262 #LDNOTE : changed original implementation below because it did not make
1263 # sense when bindtype eq 'columns' and @vals > 1.
1264 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
1266 # called often - tighten code
1267 return $_[0]->{bindtype} eq 'columns'
1268 ? map {[$_[1], $_]} @_[2 .. $#_]
1273 # Dies if any element of @bind is not in [colname => value] format
1274 # if bindtype is 'columns'.
1275 sub _assert_bindval_matches_bindtype {
1276 # my ($self, @bind) = @_;
1278 if ($self->{bindtype} eq 'columns') {
1280 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1281 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1287 sub _join_sql_clauses {
1288 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1290 if (@$clauses_aref > 1) {
1291 my $join = " " . $self->_sqlcase($logic) . " ";
1292 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1293 return ($sql, @$bind_aref);
1295 elsif (@$clauses_aref) {
1296 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1299 return (); # if no SQL, ignore @$bind_aref
1304 # Fix SQL case, if so requested
1306 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1307 # don't touch the argument ... crooked logic, but let's not change it!
1308 return $_[0]->{case} ? $_[1] : uc($_[1]);
1312 #======================================================================
1313 # DISPATCHING FROM REFKIND
1314 #======================================================================
1317 my ($self, $data) = @_;
1319 return 'UNDEF' unless defined $data;
1321 # blessed objects are treated like scalars
1322 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1324 return 'SCALAR' unless $ref;
1327 while ($ref eq 'REF') {
1329 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1333 return ($ref||'SCALAR') . ('REF' x $n_steps);
1337 my ($self, $data) = @_;
1338 my @try = ($self->_refkind($data));
1339 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1340 push @try, 'FALLBACK';
1344 sub _METHOD_FOR_refkind {
1345 my ($self, $meth_prefix, $data) = @_;
1348 for (@{$self->_try_refkind($data)}) {
1349 $method = $self->can($meth_prefix."_".$_)
1353 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1357 sub _SWITCH_refkind {
1358 my ($self, $data, $dispatch_table) = @_;
1361 for (@{$self->_try_refkind($data)}) {
1362 $coderef = $dispatch_table->{$_}
1366 puke "no dispatch entry for ".$self->_refkind($data)
1375 #======================================================================
1376 # VALUES, GENERATE, AUTOLOAD
1377 #======================================================================
1379 # LDNOTE: original code from nwiger, didn't touch code in that section
1380 # I feel the AUTOLOAD stuff should not be the default, it should
1381 # only be activated on explicit demand by user.
1385 my $data = shift || return;
1386 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1387 unless ref $data eq 'HASH';
1390 foreach my $k ( sort keys %$data ) {
1391 my $v = $data->{$k};
1392 $self->_SWITCH_refkind($v, {
1394 if ($self->{array_datatypes}) { # array datatype
1395 push @all_bind, $self->_bindtype($k, $v);
1397 else { # literal SQL with bind
1398 my ($sql, @bind) = @$v;
1399 $self->_assert_bindval_matches_bindtype(@bind);
1400 push @all_bind, @bind;
1403 ARRAYREFREF => sub { # literal SQL with bind
1404 my ($sql, @bind) = @${$v};
1405 $self->_assert_bindval_matches_bindtype(@bind);
1406 push @all_bind, @bind;
1408 SCALARREF => sub { # literal SQL without bind
1410 SCALAR_or_UNDEF => sub {
1411 push @all_bind, $self->_bindtype($k, $v);
1422 my(@sql, @sqlq, @sqlv);
1426 if ($ref eq 'HASH') {
1427 for my $k (sort keys %$_) {
1430 my $label = $self->_quote($k);
1431 if ($r eq 'ARRAY') {
1432 # literal SQL with bind
1433 my ($sql, @bind) = @$v;
1434 $self->_assert_bindval_matches_bindtype(@bind);
1435 push @sqlq, "$label = $sql";
1437 } elsif ($r eq 'SCALAR') {
1438 # literal SQL without bind
1439 push @sqlq, "$label = $$v";
1441 push @sqlq, "$label = ?";
1442 push @sqlv, $self->_bindtype($k, $v);
1445 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1446 } elsif ($ref eq 'ARRAY') {
1447 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1450 if ($r eq 'ARRAY') { # literal SQL with bind
1451 my ($sql, @bind) = @$v;
1452 $self->_assert_bindval_matches_bindtype(@bind);
1455 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1456 # embedded literal SQL
1463 push @sql, '(' . join(', ', @sqlq) . ')';
1464 } elsif ($ref eq 'SCALAR') {
1468 # strings get case twiddled
1469 push @sql, $self->_sqlcase($_);
1473 my $sql = join ' ', @sql;
1475 # this is pretty tricky
1476 # if ask for an array, return ($stmt, @bind)
1477 # otherwise, s/?/shift @sqlv/ to put it inline
1479 return ($sql, @sqlv);
1481 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1482 ref $d ? $d->[1] : $d/e;
1491 # This allows us to check for a local, then _form, attr
1493 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1494 return $self->generate($name, @_);
1505 SQL::Abstract - Generate SQL from Perl data structures
1511 my $sql = SQL::Abstract->new;
1513 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1515 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1517 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1519 my($stmt, @bind) = $sql->delete($table, \%where);
1521 # Then, use these in your DBI statements
1522 my $sth = $dbh->prepare($stmt);
1523 $sth->execute(@bind);
1525 # Just generate the WHERE clause
1526 my($stmt, @bind) = $sql->where(\%where, \@order);
1528 # Return values in the same order, for hashed queries
1529 # See PERFORMANCE section for more details
1530 my @bind = $sql->values(\%fieldvals);
1534 This module was inspired by the excellent L<DBIx::Abstract>.
1535 However, in using that module I found that what I really wanted
1536 to do was generate SQL, but still retain complete control over my
1537 statement handles and use the DBI interface. So, I set out to
1538 create an abstract SQL generation module.
1540 While based on the concepts used by L<DBIx::Abstract>, there are
1541 several important differences, especially when it comes to WHERE
1542 clauses. I have modified the concepts used to make the SQL easier
1543 to generate from Perl data structures and, IMO, more intuitive.
1544 The underlying idea is for this module to do what you mean, based
1545 on the data structures you provide it. The big advantage is that
1546 you don't have to modify your code every time your data changes,
1547 as this module figures it out.
1549 To begin with, an SQL INSERT is as easy as just specifying a hash
1550 of C<key=value> pairs:
1553 name => 'Jimbo Bobson',
1554 phone => '123-456-7890',
1555 address => '42 Sister Lane',
1556 city => 'St. Louis',
1557 state => 'Louisiana',
1560 The SQL can then be generated with this:
1562 my($stmt, @bind) = $sql->insert('people', \%data);
1564 Which would give you something like this:
1566 $stmt = "INSERT INTO people
1567 (address, city, name, phone, state)
1568 VALUES (?, ?, ?, ?, ?)";
1569 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1570 '123-456-7890', 'Louisiana');
1572 These are then used directly in your DBI code:
1574 my $sth = $dbh->prepare($stmt);
1575 $sth->execute(@bind);
1577 =head2 Inserting and Updating Arrays
1579 If your database has array types (like for example Postgres),
1580 activate the special option C<< array_datatypes => 1 >>
1581 when creating the C<SQL::Abstract> object.
1582 Then you may use an arrayref to insert and update database array types:
1584 my $sql = SQL::Abstract->new(array_datatypes => 1);
1586 planets => [qw/Mercury Venus Earth Mars/]
1589 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1593 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1595 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1598 =head2 Inserting and Updating SQL
1600 In order to apply SQL functions to elements of your C<%data> you may
1601 specify a reference to an arrayref for the given hash value. For example,
1602 if you need to execute the Oracle C<to_date> function on a value, you can
1603 say something like this:
1607 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1610 The first value in the array is the actual SQL. Any other values are
1611 optional and would be included in the bind values array. This gives
1614 my($stmt, @bind) = $sql->insert('people', \%data);
1616 $stmt = "INSERT INTO people (name, date_entered)
1617 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1618 @bind = ('Bill', '03/02/2003');
1620 An UPDATE is just as easy, all you change is the name of the function:
1622 my($stmt, @bind) = $sql->update('people', \%data);
1624 Notice that your C<%data> isn't touched; the module will generate
1625 the appropriately quirky SQL for you automatically. Usually you'll
1626 want to specify a WHERE clause for your UPDATE, though, which is
1627 where handling C<%where> hashes comes in handy...
1629 =head2 Complex where statements
1631 This module can generate pretty complicated WHERE statements
1632 easily. For example, simple C<key=value> pairs are taken to mean
1633 equality, and if you want to see if a field is within a set
1634 of values, you can use an arrayref. Let's say we wanted to
1635 SELECT some data based on this criteria:
1638 requestor => 'inna',
1639 worker => ['nwiger', 'rcwe', 'sfz'],
1640 status => { '!=', 'completed' }
1643 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1645 The above would give you something like this:
1647 $stmt = "SELECT * FROM tickets WHERE
1648 ( requestor = ? ) AND ( status != ? )
1649 AND ( worker = ? OR worker = ? OR worker = ? )";
1650 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1652 Which you could then use in DBI code like so:
1654 my $sth = $dbh->prepare($stmt);
1655 $sth->execute(@bind);
1661 The functions are simple. There's one for each major SQL operation,
1662 and a constructor you use first. The arguments are specified in a
1663 similar order to each function (table, then fields, then a where
1664 clause) to try and simplify things.
1669 =head2 new(option => 'value')
1671 The C<new()> function takes a list of options and values, and returns
1672 a new B<SQL::Abstract> object which can then be used to generate SQL
1673 through the methods below. The options accepted are:
1679 If set to 'lower', then SQL will be generated in all lowercase. By
1680 default SQL is generated in "textbook" case meaning something like:
1682 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1684 Any setting other than 'lower' is ignored.
1688 This determines what the default comparison operator is. By default
1689 it is C<=>, meaning that a hash like this:
1691 %where = (name => 'nwiger', email => 'nate@wiger.org');
1693 Will generate SQL like this:
1695 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1697 However, you may want loose comparisons by default, so if you set
1698 C<cmp> to C<like> you would get SQL such as:
1700 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1702 You can also override the comparsion on an individual basis - see
1703 the huge section on L</"WHERE CLAUSES"> at the bottom.
1705 =item sqltrue, sqlfalse
1707 Expressions for inserting boolean values within SQL statements.
1708 By default these are C<1=1> and C<1=0>. They are used
1709 by the special operators C<-in> and C<-not_in> for generating
1710 correct SQL even when the argument is an empty array (see below).
1714 This determines the default logical operator for multiple WHERE
1715 statements in arrays or hashes. If absent, the default logic is "or"
1716 for arrays, and "and" for hashes. This means that a WHERE
1720 event_date => {'>=', '2/13/99'},
1721 event_date => {'<=', '4/24/03'},
1724 will generate SQL like this:
1726 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1728 This is probably not what you want given this query, though (look
1729 at the dates). To change the "OR" to an "AND", simply specify:
1731 my $sql = SQL::Abstract->new(logic => 'and');
1733 Which will change the above C<WHERE> to:
1735 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1737 The logic can also be changed locally by inserting
1738 a modifier in front of an arrayref :
1740 @where = (-and => [event_date => {'>=', '2/13/99'},
1741 event_date => {'<=', '4/24/03'} ]);
1743 See the L</"WHERE CLAUSES"> section for explanations.
1747 This will automatically convert comparisons using the specified SQL
1748 function for both column and value. This is mostly used with an argument
1749 of C<upper> or C<lower>, so that the SQL will have the effect of
1750 case-insensitive "searches". For example, this:
1752 $sql = SQL::Abstract->new(convert => 'upper');
1753 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1755 Will turn out the following SQL:
1757 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1759 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1760 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1761 not validate this option; it will just pass through what you specify verbatim).
1765 This is a kludge because many databases suck. For example, you can't
1766 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1767 Instead, you have to use C<bind_param()>:
1769 $sth->bind_param(1, 'reg data');
1770 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1772 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1773 which loses track of which field each slot refers to. Fear not.
1775 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1776 Currently, you can specify either C<normal> (default) or C<columns>. If you
1777 specify C<columns>, you will get an array that looks like this:
1779 my $sql = SQL::Abstract->new(bindtype => 'columns');
1780 my($stmt, @bind) = $sql->insert(...);
1783 [ 'column1', 'value1' ],
1784 [ 'column2', 'value2' ],
1785 [ 'column3', 'value3' ],
1788 You can then iterate through this manually, using DBI's C<bind_param()>.
1790 $sth->prepare($stmt);
1793 my($col, $data) = @$_;
1794 if ($col eq 'details' || $col eq 'comments') {
1795 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1796 } elsif ($col eq 'image') {
1797 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1799 $sth->bind_param($i, $data);
1803 $sth->execute; # execute without @bind now
1805 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1806 Basically, the advantage is still that you don't have to care which fields
1807 are or are not included. You could wrap that above C<for> loop in a simple
1808 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1809 get a layer of abstraction over manual SQL specification.
1811 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1812 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1813 will expect the bind values in this format.
1817 This is the character that a table or column name will be quoted
1818 with. By default this is an empty string, but you could set it to
1819 the character C<`>, to generate SQL like this:
1821 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1823 Alternatively, you can supply an array ref of two items, the first being the left
1824 hand quote character, and the second the right hand quote character. For
1825 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1826 that generates SQL like this:
1828 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1830 Quoting is useful if you have tables or columns names that are reserved
1831 words in your database's SQL dialect.
1835 This is the character that separates a table and column name. It is
1836 necessary to specify this when the C<quote_char> option is selected,
1837 so that tables and column names can be individually quoted like this:
1839 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1841 =item injection_guard
1843 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1844 column name specified in a query structure. This is a safety mechanism to avoid
1845 injection attacks when mishandling user input e.g.:
1847 my %condition_as_column_value_pairs = get_values_from_user();
1848 $sqla->select( ... , \%condition_as_column_value_pairs );
1850 If the expression matches an exception is thrown. Note that literal SQL
1851 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1853 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1855 =item array_datatypes
1857 When this option is true, arrayrefs in INSERT or UPDATE are
1858 interpreted as array datatypes and are passed directly
1860 When this option is false, arrayrefs are interpreted
1861 as literal SQL, just like refs to arrayrefs
1862 (but this behavior is for backwards compatibility; when writing
1863 new queries, use the "reference to arrayref" syntax
1869 Takes a reference to a list of "special operators"
1870 to extend the syntax understood by L<SQL::Abstract>.
1871 See section L</"SPECIAL OPERATORS"> for details.
1875 Takes a reference to a list of "unary operators"
1876 to extend the syntax understood by L<SQL::Abstract>.
1877 See section L</"UNARY OPERATORS"> for details.
1883 =head2 insert($table, \@values || \%fieldvals, \%options)
1885 This is the simplest function. You simply give it a table name
1886 and either an arrayref of values or hashref of field/value pairs.
1887 It returns an SQL INSERT statement and a list of bind values.
1888 See the sections on L</"Inserting and Updating Arrays"> and
1889 L</"Inserting and Updating SQL"> for information on how to insert
1890 with those data types.
1892 The optional C<\%options> hash reference may contain additional
1893 options to generate the insert SQL. Currently supported options
1900 Takes either a scalar of raw SQL fields, or an array reference of
1901 field names, and adds on an SQL C<RETURNING> statement at the end.
1902 This allows you to return data generated by the insert statement
1903 (such as row IDs) without performing another C<SELECT> statement.
1904 Note, however, this is not part of the SQL standard and may not
1905 be supported by all database engines.
1909 =head2 update($table, \%fieldvals, \%where)
1911 This takes a table, hashref of field/value pairs, and an optional
1912 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1914 See the sections on L</"Inserting and Updating Arrays"> and
1915 L</"Inserting and Updating SQL"> for information on how to insert
1916 with those data types.
1918 =head2 select($source, $fields, $where, $order)
1920 This returns a SQL SELECT statement and associated list of bind values, as
1921 specified by the arguments :
1927 Specification of the 'FROM' part of the statement.
1928 The argument can be either a plain scalar (interpreted as a table
1929 name, will be quoted), or an arrayref (interpreted as a list
1930 of table names, joined by commas, quoted), or a scalarref
1931 (literal table name, not quoted), or a ref to an arrayref
1932 (list of literal table names, joined by commas, not quoted).
1936 Specification of the list of fields to retrieve from
1938 The argument can be either an arrayref (interpreted as a list
1939 of field names, will be joined by commas and quoted), or a
1940 plain scalar (literal SQL, not quoted).
1941 Please observe that this API is not as flexible as for
1942 the first argument C<$table>, for backwards compatibility reasons.
1946 Optional argument to specify the WHERE part of the query.
1947 The argument is most often a hashref, but can also be
1948 an arrayref or plain scalar --
1949 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1953 Optional argument to specify the ORDER BY part of the query.
1954 The argument can be a scalar, a hashref or an arrayref
1955 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1961 =head2 delete($table, \%where)
1963 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1964 It returns an SQL DELETE statement and list of bind values.
1966 =head2 where(\%where, \@order)
1968 This is used to generate just the WHERE clause. For example,
1969 if you have an arbitrary data structure and know what the
1970 rest of your SQL is going to look like, but want an easy way
1971 to produce a WHERE clause, use this. It returns an SQL WHERE
1972 clause and list of bind values.
1975 =head2 values(\%data)
1977 This just returns the values from the hash C<%data>, in the same
1978 order that would be returned from any of the other above queries.
1979 Using this allows you to markedly speed up your queries if you
1980 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1982 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1984 Warning: This is an experimental method and subject to change.
1986 This returns arbitrarily generated SQL. It's a really basic shortcut.
1987 It will return two different things, depending on return context:
1989 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1990 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1992 These would return the following:
1994 # First calling form
1995 $stmt = "CREATE TABLE test (?, ?)";
1996 @bind = (field1, field2);
1998 # Second calling form
1999 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2001 Depending on what you're trying to do, it's up to you to choose the correct
2002 format. In this example, the second form is what you would want.
2006 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2010 ALTER SESSION SET nls_date_format = 'MM/YY'
2012 You get the idea. Strings get their case twiddled, but everything
2013 else remains verbatim.
2015 =head1 WHERE CLAUSES
2019 This module uses a variation on the idea from L<DBIx::Abstract>. It
2020 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2021 module is that things in arrays are OR'ed, and things in hashes
2024 The easiest way to explain is to show lots of examples. After
2025 each C<%where> hash shown, it is assumed you used:
2027 my($stmt, @bind) = $sql->where(\%where);
2029 However, note that the C<%where> hash can be used directly in any
2030 of the other functions as well, as described above.
2032 =head2 Key-value pairs
2034 So, let's get started. To begin, a simple hash:
2038 status => 'completed'
2041 Is converted to SQL C<key = val> statements:
2043 $stmt = "WHERE user = ? AND status = ?";
2044 @bind = ('nwiger', 'completed');
2046 One common thing I end up doing is having a list of values that
2047 a field can be in. To do this, simply specify a list inside of
2052 status => ['assigned', 'in-progress', 'pending'];
2055 This simple code will create the following:
2057 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2058 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2060 A field associated to an empty arrayref will be considered a
2061 logical false and will generate 0=1.
2063 =head2 Tests for NULL values
2065 If the value part is C<undef> then this is converted to SQL <IS NULL>
2074 $stmt = "WHERE user = ? AND status IS NULL";
2077 To test if a column IS NOT NULL:
2081 status => { '!=', undef },
2084 =head2 Specific comparison operators
2086 If you want to specify a different type of operator for your comparison,
2087 you can use a hashref for a given column:
2091 status => { '!=', 'completed' }
2094 Which would generate:
2096 $stmt = "WHERE user = ? AND status != ?";
2097 @bind = ('nwiger', 'completed');
2099 To test against multiple values, just enclose the values in an arrayref:
2101 status => { '=', ['assigned', 'in-progress', 'pending'] };
2103 Which would give you:
2105 "WHERE status = ? OR status = ? OR status = ?"
2108 The hashref can also contain multiple pairs, in which case it is expanded
2109 into an C<AND> of its elements:
2113 status => { '!=', 'completed', -not_like => 'pending%' }
2116 # Or more dynamically, like from a form
2117 $where{user} = 'nwiger';
2118 $where{status}{'!='} = 'completed';
2119 $where{status}{'-not_like'} = 'pending%';
2121 # Both generate this
2122 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2123 @bind = ('nwiger', 'completed', 'pending%');
2126 To get an OR instead, you can combine it with the arrayref idea:
2130 priority => [ { '=', 2 }, { '>', 5 } ]
2133 Which would generate:
2135 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2136 @bind = ('2', '5', 'nwiger');
2138 If you want to include literal SQL (with or without bind values), just use a
2139 scalar reference or array reference as the value:
2142 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2143 date_expires => { '<' => \"now()" }
2146 Which would generate:
2148 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2149 @bind = ('11/26/2008');
2152 =head2 Logic and nesting operators
2154 In the example above,
2155 there is a subtle trap if you want to say something like
2156 this (notice the C<AND>):
2158 WHERE priority != ? AND priority != ?
2160 Because, in Perl you I<can't> do this:
2162 priority => { '!=', 2, '!=', 1 }
2164 As the second C<!=> key will obliterate the first. The solution
2165 is to use the special C<-modifier> form inside an arrayref:
2167 priority => [ -and => {'!=', 2},
2171 Normally, these would be joined by C<OR>, but the modifier tells it
2172 to use C<AND> instead. (Hint: You can use this in conjunction with the
2173 C<logic> option to C<new()> in order to change the way your queries
2174 work by default.) B<Important:> Note that the C<-modifier> goes
2175 B<INSIDE> the arrayref, as an extra first element. This will
2176 B<NOT> do what you think it might:
2178 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2180 Here is a quick list of equivalencies, since there is some overlap:
2183 status => {'!=', 'completed', 'not like', 'pending%' }
2184 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2187 status => {'=', ['assigned', 'in-progress']}
2188 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2189 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2193 =head2 Special operators : IN, BETWEEN, etc.
2195 You can also use the hashref format to compare a list of fields using the
2196 C<IN> comparison operator, by specifying the list as an arrayref:
2199 status => 'completed',
2200 reportid => { -in => [567, 2335, 2] }
2203 Which would generate:
2205 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2206 @bind = ('completed', '567', '2335', '2');
2208 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2211 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2212 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2213 'sqltrue' (by default : C<1=1>).
2215 In addition to the array you can supply a chunk of literal sql or
2216 literal sql with bind:
2219 customer => { -in => \[
2220 'SELECT cust_id FROM cust WHERE balance > ?',
2223 status => { -in => \'SELECT status_codes FROM states' },
2229 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2230 AND status IN ( SELECT status_codes FROM states )
2236 Another pair of operators is C<-between> and C<-not_between>,
2237 used with an arrayref of two values:
2241 completion_date => {
2242 -not_between => ['2002-10-01', '2003-02-06']
2248 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2250 Just like with C<-in> all plausible combinations of literal SQL
2254 start0 => { -between => [ 1, 2 ] },
2255 start1 => { -between => \["? AND ?", 1, 2] },
2256 start2 => { -between => \"lower(x) AND upper(y)" },
2257 start3 => { -between => [
2259 \["upper(?)", 'stuff' ],
2266 ( start0 BETWEEN ? AND ? )
2267 AND ( start1 BETWEEN ? AND ? )
2268 AND ( start2 BETWEEN lower(x) AND upper(y) )
2269 AND ( start3 BETWEEN lower(x) AND upper(?) )
2271 @bind = (1, 2, 1, 2, 'stuff');
2274 These are the two builtin "special operators"; but the
2275 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2277 =head2 Unary operators: bool
2279 If you wish to test against boolean columns or functions within your
2280 database you can use the C<-bool> and C<-not_bool> operators. For
2281 example to test the column C<is_user> being true and the column
2282 C<is_enabled> being false you would use:-
2286 -not_bool => 'is_enabled',
2291 WHERE is_user AND NOT is_enabled
2293 If a more complex combination is required, testing more conditions,
2294 then you should use the and/or operators:-
2301 -not_bool => 'four',
2307 WHERE one AND two AND three AND NOT four
2310 =head2 Nested conditions, -and/-or prefixes
2312 So far, we've seen how multiple conditions are joined with a top-level
2313 C<AND>. We can change this by putting the different conditions we want in
2314 hashes and then putting those hashes in an array. For example:
2319 status => { -like => ['pending%', 'dispatched'] },
2323 status => 'unassigned',
2327 This data structure would create the following:
2329 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2330 OR ( user = ? AND status = ? ) )";
2331 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2334 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2335 to change the logic inside :
2341 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2342 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2349 WHERE ( user = ? AND (
2350 ( workhrs > ? AND geo = ? )
2351 OR ( workhrs < ? OR geo = ? )
2354 =head3 Algebraic inconsistency, for historical reasons
2356 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2357 operator goes C<outside> of the nested structure; whereas when connecting
2358 several constraints on one column, the C<-and> operator goes
2359 C<inside> the arrayref. Here is an example combining both features :
2362 -and => [a => 1, b => 2],
2363 -or => [c => 3, d => 4],
2364 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2369 WHERE ( ( ( a = ? AND b = ? )
2370 OR ( c = ? OR d = ? )
2371 OR ( e LIKE ? AND e LIKE ? ) ) )
2373 This difference in syntax is unfortunate but must be preserved for
2374 historical reasons. So be careful : the two examples below would
2375 seem algebraically equivalent, but they are not
2377 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2378 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2380 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2381 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2384 =head2 Literal SQL and value type operators
2386 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2387 side" is a column name and the "right side" is a value (normally rendered as
2388 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2389 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2390 alter this behavior. There are several ways of doing so.
2394 This is a virtual operator that signals the string to its right side is an
2395 identifier (a column name) and not a value. For example to compare two
2396 columns you would write:
2399 priority => { '<', 2 },
2400 requestor => { -ident => 'submitter' },
2405 $stmt = "WHERE priority < ? AND requestor = submitter";
2408 If you are maintaining legacy code you may see a different construct as
2409 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2414 This is a virtual operator that signals that the construct to its right side
2415 is a value to be passed to DBI. This is for example necessary when you want
2416 to write a where clause against an array (for RDBMS that support such
2417 datatypes). For example:
2420 array => { -value => [1, 2, 3] }
2425 $stmt = 'WHERE array = ?';
2426 @bind = ([1, 2, 3]);
2428 Note that if you were to simply say:
2434 the result would porbably be not what you wanted:
2436 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2441 Finally, sometimes only literal SQL will do. To include a random snippet
2442 of SQL verbatim, you specify it as a scalar reference. Consider this only
2443 as a last resort. Usually there is a better way. For example:
2446 priority => { '<', 2 },
2447 requestor => { -in => \'(SELECT name FROM hitmen)' },
2452 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2455 Note that in this example, you only get one bind parameter back, since
2456 the verbatim SQL is passed as part of the statement.
2460 Never use untrusted input as a literal SQL argument - this is a massive
2461 security risk (there is no way to check literal snippets for SQL
2462 injections and other nastyness). If you need to deal with untrusted input
2463 use literal SQL with placeholders as described next.
2465 =head3 Literal SQL with placeholders and bind values (subqueries)
2467 If the literal SQL to be inserted has placeholders and bind values,
2468 use a reference to an arrayref (yes this is a double reference --
2469 not so common, but perfectly legal Perl). For example, to find a date
2470 in Postgres you can use something like this:
2473 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2478 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2481 Note that you must pass the bind values in the same format as they are returned
2482 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2483 provide the bind values in the C<< [ column_meta => value ] >> format, where
2484 C<column_meta> is an opaque scalar value; most commonly the column name, but
2485 you can use any scalar value (including references and blessed references),
2486 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2487 to C<columns> the above example will look like:
2490 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2493 Literal SQL is especially useful for nesting parenthesized clauses in the
2494 main SQL query. Here is a first example :
2496 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2500 bar => \["IN ($sub_stmt)" => @sub_bind],
2505 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2506 WHERE c2 < ? AND c3 LIKE ?))";
2507 @bind = (1234, 100, "foo%");
2509 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2510 are expressed in the same way. Of course the C<$sub_stmt> and
2511 its associated bind values can be generated through a former call
2514 my ($sub_stmt, @sub_bind)
2515 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2516 c3 => {-like => "foo%"}});
2519 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2522 In the examples above, the subquery was used as an operator on a column;
2523 but the same principle also applies for a clause within the main C<%where>
2524 hash, like an EXISTS subquery :
2526 my ($sub_stmt, @sub_bind)
2527 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2528 my %where = ( -and => [
2530 \["EXISTS ($sub_stmt)" => @sub_bind],
2535 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2536 WHERE c1 = ? AND c2 > t0.c0))";
2540 Observe that the condition on C<c2> in the subquery refers to
2541 column C<t0.c0> of the main query : this is I<not> a bind
2542 value, so we have to express it through a scalar ref.
2543 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2544 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2545 what we wanted here.
2547 Finally, here is an example where a subquery is used
2548 for expressing unary negation:
2550 my ($sub_stmt, @sub_bind)
2551 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2552 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2554 lname => {like => '%son%'},
2555 \["NOT ($sub_stmt)" => @sub_bind],
2560 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2561 @bind = ('%son%', 10, 20)
2563 =head3 Deprecated usage of Literal SQL
2565 Below are some examples of archaic use of literal SQL. It is shown only as
2566 reference for those who deal with legacy code. Each example has a much
2567 better, cleaner and safer alternative that users should opt for in new code.
2573 my %where = ( requestor => \'IS NOT NULL' )
2575 $stmt = "WHERE requestor IS NOT NULL"
2577 This used to be the way of generating NULL comparisons, before the handling
2578 of C<undef> got formalized. For new code please use the superior syntax as
2579 described in L</Tests for NULL values>.
2583 my %where = ( requestor => \'= submitter' )
2585 $stmt = "WHERE requestor = submitter"
2587 This used to be the only way to compare columns. Use the superior L</-ident>
2588 method for all new code. For example an identifier declared in such a way
2589 will be properly quoted if L</quote_char> is properly set, while the legacy
2590 form will remain as supplied.
2594 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2596 $stmt = "WHERE completed > ? AND is_ready"
2597 @bind = ('2012-12-21')
2599 Using an empty string literal used to be the only way to express a boolean.
2600 For all new code please use the much more readable
2601 L<-bool|/Unary operators: bool> operator.
2607 These pages could go on for a while, since the nesting of the data
2608 structures this module can handle are pretty much unlimited (the
2609 module implements the C<WHERE> expansion as a recursive function
2610 internally). Your best bet is to "play around" with the module a
2611 little to see how the data structures behave, and choose the best
2612 format for your data based on that.
2614 And of course, all the values above will probably be replaced with
2615 variables gotten from forms or the command line. After all, if you
2616 knew everything ahead of time, you wouldn't have to worry about
2617 dynamically-generating SQL and could just hardwire it into your
2620 =head1 ORDER BY CLAUSES
2622 Some functions take an order by clause. This can either be a scalar (just a
2623 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2624 or an array of either of the two previous forms. Examples:
2626 Given | Will Generate
2627 ----------------------------------------------------------
2629 \'colA DESC' | ORDER BY colA DESC
2631 'colA' | ORDER BY colA
2633 [qw/colA colB/] | ORDER BY colA, colB
2635 {-asc => 'colA'} | ORDER BY colA ASC
2637 {-desc => 'colB'} | ORDER BY colB DESC
2639 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2641 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2644 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2645 { -desc => [qw/colB/], | colC ASC, colD ASC
2646 { -asc => [qw/colC colD/],|
2648 ===========================================================
2652 =head1 SPECIAL OPERATORS
2654 my $sqlmaker = SQL::Abstract->new(special_ops => [
2658 my ($self, $field, $op, $arg) = @_;
2664 handler => 'method_name',
2668 A "special operator" is a SQL syntactic clause that can be
2669 applied to a field, instead of a usual binary operator.
2672 WHERE field IN (?, ?, ?)
2673 WHERE field BETWEEN ? AND ?
2674 WHERE MATCH(field) AGAINST (?, ?)
2676 Special operators IN and BETWEEN are fairly standard and therefore
2677 are builtin within C<SQL::Abstract> (as the overridable methods
2678 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2679 like the MATCH .. AGAINST example above which is specific to MySQL,
2680 you can write your own operator handlers - supply a C<special_ops>
2681 argument to the C<new> method. That argument takes an arrayref of
2682 operator definitions; each operator definition is a hashref with two
2689 the regular expression to match the operator
2693 Either a coderef or a plain scalar method name. In both cases
2694 the expected return is C<< ($sql, @bind) >>.
2696 When supplied with a method name, it is simply called on the
2697 L<SQL::Abstract/> object as:
2699 $self->$method_name ($field, $op, $arg)
2703 $op is the part that matched the handler regex
2704 $field is the LHS of the operator
2707 When supplied with a coderef, it is called as:
2709 $coderef->($self, $field, $op, $arg)
2714 For example, here is an implementation
2715 of the MATCH .. AGAINST syntax for MySQL
2717 my $sqlmaker = SQL::Abstract->new(special_ops => [
2719 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2720 {regex => qr/^match$/i,
2722 my ($self, $field, $op, $arg) = @_;
2723 $arg = [$arg] if not ref $arg;
2724 my $label = $self->_quote($field);
2725 my ($placeholder) = $self->_convert('?');
2726 my $placeholders = join ", ", (($placeholder) x @$arg);
2727 my $sql = $self->_sqlcase('match') . " ($label) "
2728 . $self->_sqlcase('against') . " ($placeholders) ";
2729 my @bind = $self->_bindtype($field, @$arg);
2730 return ($sql, @bind);
2737 =head1 UNARY OPERATORS
2739 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2743 my ($self, $op, $arg) = @_;
2749 handler => 'method_name',
2753 A "unary operator" is a SQL syntactic clause that can be
2754 applied to a field - the operator goes before the field
2756 You can write your own operator handlers - supply a C<unary_ops>
2757 argument to the C<new> method. That argument takes an arrayref of
2758 operator definitions; each operator definition is a hashref with two
2765 the regular expression to match the operator
2769 Either a coderef or a plain scalar method name. In both cases
2770 the expected return is C<< $sql >>.
2772 When supplied with a method name, it is simply called on the
2773 L<SQL::Abstract/> object as:
2775 $self->$method_name ($op, $arg)
2779 $op is the part that matched the handler regex
2780 $arg is the RHS or argument of the operator
2782 When supplied with a coderef, it is called as:
2784 $coderef->($self, $op, $arg)
2792 Thanks to some benchmarking by Mark Stosberg, it turns out that
2793 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2794 I must admit this wasn't an intentional design issue, but it's a
2795 byproduct of the fact that you get to control your C<DBI> handles
2798 To maximize performance, use a code snippet like the following:
2800 # prepare a statement handle using the first row
2801 # and then reuse it for the rest of the rows
2803 for my $href (@array_of_hashrefs) {
2804 $stmt ||= $sql->insert('table', $href);
2805 $sth ||= $dbh->prepare($stmt);
2806 $sth->execute($sql->values($href));
2809 The reason this works is because the keys in your C<$href> are sorted
2810 internally by B<SQL::Abstract>. Thus, as long as your data retains
2811 the same structure, you only have to generate the SQL the first time
2812 around. On subsequent queries, simply use the C<values> function provided
2813 by this module to return your values in the correct order.
2815 However this depends on the values having the same type - if, for
2816 example, the values of a where clause may either have values
2817 (resulting in sql of the form C<column = ?> with a single bind
2818 value), or alternatively the values might be C<undef> (resulting in
2819 sql of the form C<column IS NULL> with no bind value) then the
2820 caching technique suggested will not work.
2824 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2825 really like this part (I do, at least). Building up a complex query
2826 can be as simple as the following:
2830 use CGI::FormBuilder;
2833 my $form = CGI::FormBuilder->new(...);
2834 my $sql = SQL::Abstract->new;
2836 if ($form->submitted) {
2837 my $field = $form->field;
2838 my $id = delete $field->{id};
2839 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2842 Of course, you would still have to connect using C<DBI> to run the
2843 query, but the point is that if you make your form look like your
2844 table, the actual query script can be extremely simplistic.
2846 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2847 a fast interface to returning and formatting data. I frequently
2848 use these three modules together to write complex database query
2849 apps in under 50 lines.
2855 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2857 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2863 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2864 Great care has been taken to preserve the I<published> behavior
2865 documented in previous versions in the 1.* family; however,
2866 some features that were previously undocumented, or behaved
2867 differently from the documentation, had to be changed in order
2868 to clarify the semantics. Hence, client code that was relying
2869 on some dark areas of C<SQL::Abstract> v1.*
2870 B<might behave differently> in v1.50.
2872 The main changes are :
2878 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2882 support for the { operator => \"..." } construct (to embed literal SQL)
2886 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2890 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2894 defensive programming : check arguments
2898 fixed bug with global logic, which was previously implemented
2899 through global variables yielding side-effects. Prior versions would
2900 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2901 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2902 Now this is interpreted
2903 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2908 fixed semantics of _bindtype on array args
2912 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2913 we just avoid shifting arrays within that tree.
2917 dropped the C<_modlogic> function
2921 =head1 ACKNOWLEDGEMENTS
2923 There are a number of individuals that have really helped out with
2924 this module. Unfortunately, most of them submitted bugs via CPAN
2925 so I have no idea who they are! But the people I do know are:
2927 Ash Berlin (order_by hash term support)
2928 Matt Trout (DBIx::Class support)
2929 Mark Stosberg (benchmarking)
2930 Chas Owens (initial "IN" operator support)
2931 Philip Collins (per-field SQL functions)
2932 Eric Kolve (hashref "AND" support)
2933 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2934 Dan Kubb (support for "quote_char" and "name_sep")
2935 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2936 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2937 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2938 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2939 Oliver Charles (support for "RETURNING" after "INSERT")
2945 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2949 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2951 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2953 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2954 While not an official support venue, C<DBIx::Class> makes heavy use of
2955 C<SQL::Abstract>, and as such list members there are very familiar with
2956 how to create queries.
2960 This module is free software; you may copy this under the same
2961 terms as perl itself (either the GNU General Public License or
2962 the Artistic License)