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)
11 use List::Util qw/first/;
13 #======================================================================
15 #======================================================================
17 our $VERSION = '1.49_01';
21 # special operators (-in, -between). May be extended/overridden by user.
22 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
23 my @BUILTIN_SPECIAL_OPS = (
24 {regex => qr/^(not )?between$/i, handler => \&_where_field_BETWEEN},
25 {regex => qr/^(not )?in$/i, handler => \&_where_field_IN},
28 #======================================================================
29 # DEBUGGING AND ERROR REPORTING
30 #======================================================================
33 return unless $_[0]->{debug}; shift; # a little faster
34 my $func = (caller(1))[3];
35 warn "[$func] ", @_, "\n";
39 my($func) = (caller(1))[3];
40 carp "[$func] Warning: ", @_;
44 my($func) = (caller(1))[3];
45 croak "[$func] Fatal: ", @_;
49 #======================================================================
51 #======================================================================
55 my $class = ref($self) || $self;
56 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
58 # choose our case by keeping an option around
59 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
61 # default logic for interpreting arrayrefs
62 $opt{logic} = uc $opt{logic} || 'OR';
64 # how to return bind vars
65 # LDNOTE: changed nwiger code : why this 'delete' ??
66 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
67 $opt{bindtype} ||= 'normal';
69 # default comparison is "=", but can be overridden
72 # try to recognize which are the 'equality' and 'unequality' ops
73 # (temporary quickfix, should go through a more seasoned API)
74 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
75 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
78 $opt{sqltrue} ||= '1=1';
79 $opt{sqlfalse} ||= '0=1';
82 $opt{special_ops} ||= [];
83 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
85 return bless \%opt, $class;
90 #======================================================================
92 #======================================================================
96 my $table = $self->_table(shift);
97 my $data = shift || return;
99 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
100 my ($sql, @bind) = $self->$method($data);
101 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
102 return wantarray ? ($sql, @bind) : $sql;
105 sub _insert_HASHREF { # explicit list of fields and then values
106 my ($self, $data) = @_;
108 my @fields = sort keys %$data;
111 { # get values (need temporary override of bindtype to avoid an error)
112 local $self->{bindtype} = 'normal';
113 ($sql, @bind) = $self->_insert_ARRAYREF([@{$data}{@fields}]);
116 # if necessary, transform values according to 'bindtype'
117 if ($self->{bindtype} eq 'columns') {
118 for my $i (0 .. $#fields) {
119 ($bind[$i]) = $self->_bindtype($fields[$i], $bind[$i]);
124 $_ = $self->_quote($_) foreach @fields;
125 $sql = "( ".join(", ", @fields).") ".$sql;
127 return ($sql, @bind);
130 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
131 my ($self, $data) = @_;
133 # no names (arrayref) so can't generate bindtype
134 $self->{bindtype} ne 'columns'
135 or belch "can't do 'columns' bindtype when called with arrayref";
137 my (@values, @all_bind);
140 $self->_SWITCH_refkind($v, {
143 if ($self->{array_datatypes}) { # if array datatype are activated
146 else { # else literal SQL with bind
147 my ($sql, @bind) = @$v;
149 push @all_bind, @bind;
153 ARRAYREFREF => sub { # literal SQL with bind
154 my ($sql, @bind) = @${$v};
156 push @all_bind, @bind;
159 # THINK : anything useful to do with a HASHREF ?
161 SCALARREF => sub { # literal SQL without bind
165 SCALAR_or_UNDEF => sub {
174 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
175 return ($sql, @all_bind);
179 sub _insert_ARRAYREFREF { # literal SQL with bind
180 my ($self, $data) = @_;
185 sub _insert_SCALARREF { # literal SQL without bind
186 my ($self, $data) = @_;
193 #======================================================================
195 #======================================================================
200 my $table = $self->_table(shift);
201 my $data = shift || return;
204 # first build the 'SET' part of the sql statement
205 my (@set, @all_bind);
206 puke "Unsupported data type specified to \$sql->update"
207 unless ref $data eq 'HASH';
209 for my $k (sort keys %$data) {
212 my $label = $self->_quote($k);
214 $self->_SWITCH_refkind($v, {
216 if ($self->{array_datatypes}) { # array datatype
217 push @set, "$label = ?";
218 push @all_bind, $self->_bindtype($k, $v);
220 else { # literal SQL with bind
221 my ($sql, @bind) = @$v;
222 push @set, "$label = $sql";
223 push @all_bind, $self->_bindtype($k, @bind);
226 ARRAYREFREF => sub { # literal SQL with bind
227 my ($sql, @bind) = @${$v};
228 push @set, "$label = $sql";
229 push @all_bind, $self->_bindtype($k, @bind);
231 SCALARREF => sub { # literal SQL without bind
232 push @set, "$label = $$v";
234 SCALAR_or_UNDEF => sub {
235 push @set, "$label = ?";
236 push @all_bind, $self->_bindtype($k, $v);
242 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
246 my($where_sql, @where_bind) = $self->where($where);
248 push @all_bind, @where_bind;
251 return wantarray ? ($sql, @all_bind) : $sql;
257 #======================================================================
259 #======================================================================
264 my $table = $self->_table(shift);
265 my $fields = shift || '*';
269 my($where_sql, @bind) = $self->where($where, $order);
271 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
273 my $sql = join(' ', $self->_sqlcase('select'), $f,
274 $self->_sqlcase('from'), $table)
277 return wantarray ? ($sql, @bind) : $sql;
280 #======================================================================
282 #======================================================================
287 my $table = $self->_table(shift);
291 my($where_sql, @bind) = $self->where($where);
292 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
294 return wantarray ? ($sql, @bind) : $sql;
298 #======================================================================
300 #======================================================================
304 # Finally, a separate routine just to handle WHERE clauses
306 my ($self, $where, $order) = @_;
309 my ($sql, @bind) = $self->_recurse_where($where);
310 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
314 $sql .= $self->_order_by($order);
317 return wantarray ? ($sql, @bind) : $sql;
322 my ($self, $where, $logic) = @_;
324 # dispatch on appropriate method according to refkind of $where
325 my $method = $self->_METHOD_FOR_refkind("_where", $where);
326 $self->$method($where, $logic);
331 #======================================================================
332 # WHERE: top-level ARRAYREF
333 #======================================================================
336 sub _where_ARRAYREF {
337 my ($self, $where, $logic) = @_;
339 $logic = uc($logic || $self->{logic});
340 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
342 my @clauses = @$where;
344 # if the array starts with [-and|or => ...], recurse with that logic
345 my $first = $clauses[0] || '';
346 if ($first =~ /^-(and|or)/i) {
349 return $self->_where_ARRAYREF(\@clauses, $logic);
353 my (@sql_clauses, @all_bind);
355 # need to use while() so can shift() for pairs
356 while (my $el = shift @clauses) {
358 # switch according to kind of $el and get corresponding ($sql, @bind)
359 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
361 # skip empty elements, otherwise get invalid trailing AND stuff
362 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
364 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
365 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
366 # side-effect: the first hashref within an array would change
367 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
368 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
369 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
371 SCALARREF => sub { ($$el); },
373 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
374 $self->_recurse_where({$el => shift(@clauses)})},
376 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
380 push @sql_clauses, $sql;
381 push @all_bind, @bind;
385 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
390 #======================================================================
391 # WHERE: top-level HASHREF
392 #======================================================================
395 my ($self, $where) = @_;
396 my (@sql_clauses, @all_bind);
398 # LDNOTE : don't really know why we need to sort keys
399 for my $k (sort keys %$where) {
400 my $v = $where->{$k};
402 # ($k => $v) is either a special op or a regular hashpair
403 my ($sql, @bind) = ($k =~ /^-(.+)/) ? $self->_where_op_in_hash($1, $v)
405 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
406 $self->$method($k, $v);
409 push @sql_clauses, $sql;
410 push @all_bind, @bind;
413 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
417 sub _where_op_in_hash {
418 my ($self, $op, $v) = @_;
420 $op =~ /^(AND|OR|NEST)[_\d]*/i
421 or puke "unknown operator: -$op";
422 $op = uc($1); # uppercase, remove trailing digits
423 $self->_debug("OP(-$op) within hashref, recursing...");
425 $self->_SWITCH_refkind($v, {
428 # LDNOTE : should deprecate {-or => [...]} and {-and => [...]}
429 # because they are misleading; the only proper way would be
430 # -nest => [-or => ...], -nest => [-and ...]
431 return $self->_where_ARRAYREF($v, $op eq 'NEST' ? '' : $op);
436 belch "-or => {...} should be -nest => [...]";
437 return $self->_where_ARRAYREF([%$v], 'OR');
440 return $self->_where_HASHREF($v);
444 SCALARREF => sub { # literal SQL
446 or puke "-$op => \\\$scalar not supported, use -nest => ...";
450 ARRAYREFREF => sub { # literal SQL
452 or puke "-$op => \\[..] not supported, use -nest => ...";
456 SCALAR => sub { # permissively interpreted as SQL
458 or puke "-$op => 'scalar' not supported, use -nest => \\'scalar'";
459 belch "literal SQL should be -nest => \\'scalar' "
460 . "instead of -nest => 'scalar' ";
465 puke "-$op => undef not supported";
471 sub _where_hashpair_ARRAYREF {
472 my ($self, $k, $v) = @_;
475 my @v = @$v; # need copy because of shift below
476 $self->_debug("ARRAY($k) means distribute over elements");
478 # put apart first element if it is an operator (-and, -or)
479 my $op = $v[0] =~ /^-/ ? shift @v : undef;
480 $self->_debug("OP($op) reinjected into the distributed array") if $op;
482 my @distributed = map { {$k => $_} } @v;
483 unshift @distributed, $op if $op;
485 return $self->_recurse_where(\@distributed);
488 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
489 $self->_debug("empty ARRAY($k) means 0=1");
490 return ($self->{sqlfalse});
494 sub _where_hashpair_HASHREF {
495 my ($self, $k, $v) = @_;
497 my (@all_sql, @all_bind);
499 for my $op (sort keys %$v) {
502 # put the operator in canonical form
503 $op =~ s/^-//; # remove initial dash
504 $op =~ tr/_/ /; # underscores become spaces
505 $op =~ s/^\s+//; # no initial space
506 $op =~ s/\s+$//; # no final space
507 $op =~ s/\s+/ /; # multiple spaces become one
511 # CASE: special operators like -in or -between
512 my $special_op = first {$op =~ $_->{regex}} @{$self->{special_ops}};
514 ($sql, @bind) = $special_op->{handler}->($self, $k, $op, $val);
517 # CASE: col => {op => \@vals}
518 elsif (ref $val eq 'ARRAY') {
519 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
522 # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
523 elsif (! defined($val)) {
524 my $is = ($op =~ $self->{equality_op}) ? 'is' :
525 ($op =~ $self->{inequality_op}) ? 'is not' :
526 puke "unexpected operator '$op' with undef operand";
527 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
530 # CASE: col => {op => $scalar}
532 $sql = join ' ', $self->_convert($self->_quote($k)),
533 $self->_sqlcase($op),
534 $self->_convert('?');
535 @bind = $self->_bindtype($k, $val);
539 push @all_bind, @bind;
542 return $self->_join_sql_clauses('and', \@all_sql, \@all_bind);
547 sub _where_field_op_ARRAYREF {
548 my ($self, $k, $op, $vals) = @_;
551 $self->_debug("ARRAY($vals) means multiple elements: [ @$vals ]");
555 # LDNOTE : change the distribution logic when
556 # $op =~ $self->{inequality_op}, because of Morgan laws :
557 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
558 # WHERE field != 22 OR field != 33 : the user probably means
559 # WHERE field != 22 AND field != 33.
560 my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
562 # distribute $op over each member of @$vals
563 return $self->_recurse_where([map { {$k => {$op, $_}} } @$vals], $logic);
567 # try to DWIM on equality operators
568 # LDNOTE : not 100% sure this is the correct thing to do ...
569 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
570 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
573 puke "operator '$op' applied on an empty array (field '$k')";
578 sub _where_hashpair_SCALARREF {
579 my ($self, $k, $v) = @_;
580 $self->_debug("SCALAR($k) means literal SQL: $$v");
581 my $sql = $self->_quote($k) . " " . $$v;
585 sub _where_hashpair_ARRAYREFREF {
586 my ($self, $k, $v) = @_;
587 $self->_debug("REF($k) means literal SQL: @${$v}");
588 my ($sql, @bind) = @${$v};
589 $sql = $self->_quote($k) . " " . $sql;
590 @bind = $self->_bindtype($k, @bind);
591 return ($sql, @bind );
594 sub _where_hashpair_SCALAR {
595 my ($self, $k, $v) = @_;
596 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
597 my $sql = join ' ', $self->_convert($self->_quote($k)),
598 $self->_sqlcase($self->{cmp}),
599 $self->_convert('?');
600 my @bind = $self->_bindtype($k, $v);
601 return ( $sql, @bind);
605 sub _where_hashpair_UNDEF {
606 my ($self, $k, $v) = @_;
607 $self->_debug("UNDEF($k) means IS NULL");
608 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
612 #======================================================================
613 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
614 #======================================================================
617 sub _where_SCALARREF {
618 my ($self, $where) = @_;
621 $self->_debug("SCALAR(*top) means literal SQL: $$where");
627 my ($self, $where) = @_;
630 $self->_debug("NOREF(*top) means literal SQL: $where");
641 #======================================================================
642 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
643 #======================================================================
646 sub _where_field_BETWEEN {
647 my ($self, $k, $op, $vals) = @_;
649 ref $vals eq 'ARRAY' && @$vals == 2
650 or puke "special op 'between' requires an arrayref of two values";
652 my ($label) = $self->_convert($self->_quote($k));
653 my ($placeholder) = $self->_convert('?');
654 my $and = $self->_sqlcase('and');
655 $op = $self->_sqlcase($op);
657 my $sql = "( $label $op $placeholder $and $placeholder )";
658 my @bind = $self->_bindtype($k, @$vals);
663 sub _where_field_IN {
664 my ($self, $k, $op, $vals) = @_;
666 # backwards compatibility : if scalar, force into an arrayref
667 $vals = [$vals] if defined $vals && ! ref $vals;
670 or puke "special op 'in' requires an arrayref";
672 my ($label) = $self->_convert($self->_quote($k));
673 my ($placeholder) = $self->_convert('?');
674 my $and = $self->_sqlcase('and');
675 $op = $self->_sqlcase($op);
677 if (@$vals) { # nonempty list
678 my $placeholders = join ", ", (($placeholder) x @$vals);
679 my $sql = "$label $op ( $placeholders )";
680 my @bind = $self->_bindtype($k, @$vals);
682 return ($sql, @bind);
684 else { # empty list : some databases won't understand "IN ()", so DWIM
685 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
695 #======================================================================
697 #======================================================================
700 my ($self, $arg) = @_;
702 # construct list of ordering instructions
703 my @order = $self->_SWITCH_refkind($arg, {
706 map {$self->_SWITCH_refkind($_, {
707 SCALAR => sub {$self->_quote($_)},
708 SCALARREF => sub {$$_}, # literal SQL, no quoting
709 HASHREF => sub {$self->_order_by_hash($_)}
713 SCALAR => sub {$self->_quote($arg)},
715 SCALARREF => sub {$$arg}, # literal SQL, no quoting
716 HASHREF => sub {$self->_order_by_hash($arg)},
721 my $order = join ', ', @order;
722 return $order ? $self->_sqlcase(' order by')." $order" : '';
727 my ($self, $hash) = @_;
729 # get first pair in hash
730 my ($key, $val) = each %$hash;
732 # check if one pair was found and no other pair in hash
733 $key && !(each %$hash)
734 or puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
736 my ($order) = ($key =~ /^-(desc|asc)/i)
737 or puke "invalid key in _order_by hash : $key";
739 return $self->_quote($val) ." ". $self->_sqlcase($order);
744 #======================================================================
745 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
746 #======================================================================
751 $self->_SWITCH_refkind($from, {
752 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
753 SCALAR => sub {$self->_quote($from)},
754 SCALARREF => sub {$$from},
755 ARRAYREFREF => sub {join ', ', @$from;},
760 #======================================================================
762 #======================================================================
768 $label or puke "can't quote an empty label";
770 # left and right quote characters
771 my ($ql, $qr, @other) = $self->_SWITCH_refkind($self->{quote_char}, {
772 SCALAR => sub {($self->{quote_char}, $self->{quote_char})},
773 ARRAYREF => sub {@{$self->{quote_char}}},
777 or puke "quote_char must be an arrayref of 2 values";
779 # no quoting if no quoting chars
780 $ql or return $label;
782 # no quoting for literal SQL
783 return $$label if ref($label) eq 'SCALAR';
785 # separate table / column (if applicable)
786 my $sep = $self->{name_sep} || '';
787 my @to_quote = $sep ? split /\Q$sep\E/, $label : ($label);
789 # do the quoting, except for "*" or for `table`.*
790 my @quoted = map { $_ eq '*' ? $_: $ql.$_.$qr} @to_quote;
792 # reassemble and return.
793 return join $sep, @quoted;
797 # Conversion, if applicable
799 my ($self, $arg) = @_;
801 # LDNOTE : modified the previous implementation below because
802 # it was not consistent : the first "return" is always an array,
803 # the second "return" is context-dependent. Anyway, _convert
804 # seems always used with just a single argument, so make it a
806 # return @_ unless $self->{convert};
807 # my $conv = $self->_sqlcase($self->{convert});
808 # my @ret = map { $conv.'('.$_.')' } @_;
809 # return wantarray ? @ret : $ret[0];
810 if ($self->{convert}) {
811 my $conv = $self->_sqlcase($self->{convert});
812 $arg = $conv.'('.$arg.')';
820 my($col, @vals) = @_;
822 #LDNOTE : changed original implementation below because it did not make
823 # sense when bindtype eq 'columns' and @vals > 1.
824 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
826 return $self->{bindtype} eq 'columns' ? map {[$col, $_]} @vals : @vals;
829 sub _join_sql_clauses {
830 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
832 if (@$clauses_aref > 1) {
833 my $join = " " . $self->_sqlcase($logic) . " ";
834 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
835 return ($sql, @$bind_aref);
837 elsif (@$clauses_aref) {
838 return ($clauses_aref->[0], @$bind_aref); # no parentheses
841 return (); # if no SQL, ignore @$bind_aref
846 # Fix SQL case, if so requested
850 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
851 # don't touch the argument ... crooked logic, but let's not change it!
852 return $self->{case} ? $_[0] : uc($_[0]);
856 #======================================================================
857 # DISPATCHING FROM REFKIND
858 #======================================================================
861 my ($self, $data) = @_;
865 # $suffix = 'REF' x (length of ref chain, i. e. \\[] is REFREFREF)
869 last if $ref ne 'REF';
873 return $ref ? $ref.$suffix :
874 defined $data ? 'SCALAR' :
879 my ($self, $data) = @_;
880 my @try = ($self->_refkind($data));
881 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
882 push @try, 'FALLBACK';
886 sub _METHOD_FOR_refkind {
887 my ($self, $meth_prefix, $data) = @_;
888 my $method = first {$_} map {$self->can($meth_prefix."_".$_)}
889 $self->_try_refkind($data)
890 or puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
895 sub _SWITCH_refkind {
896 my ($self, $data, $dispatch_table) = @_;
898 my $coderef = first {$_} map {$dispatch_table->{$_}}
899 $self->_try_refkind($data)
900 or puke "no dispatch entry for ".$self->_refkind($data);
907 #======================================================================
908 # VALUES, GENERATE, AUTOLOAD
909 #======================================================================
911 # LDNOTE: original code from nwiger, didn't touch code in that section
912 # I feel the AUTOLOAD stuff should not be the default, it should
913 # only be activated on explicit demand by user.
917 my $data = shift || return;
918 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
919 unless ref $data eq 'HASH';
920 return map { $self->_bindtype($_, $data->{$_}) } sort keys %$data;
926 my(@sql, @sqlq, @sqlv);
930 if ($ref eq 'HASH') {
931 for my $k (sort keys %$_) {
934 my $label = $self->_quote($k);
936 # SQL included for values
938 my $sql = shift @bind;
939 push @sqlq, "$label = $sql";
940 push @sqlv, $self->_bindtype($k, @bind);
941 } elsif ($r eq 'SCALAR') {
942 # embedded literal SQL
943 push @sqlq, "$label = $$v";
945 push @sqlq, "$label = ?";
946 push @sqlv, $self->_bindtype($k, $v);
949 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
950 } elsif ($ref eq 'ARRAY') {
951 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
956 push @sqlq, shift @val;
958 } elsif ($r eq 'SCALAR') {
959 # embedded literal SQL
966 push @sql, '(' . join(', ', @sqlq) . ')';
967 } elsif ($ref eq 'SCALAR') {
971 # strings get case twiddled
972 push @sql, $self->_sqlcase($_);
976 my $sql = join ' ', @sql;
978 # this is pretty tricky
979 # if ask for an array, return ($stmt, @bind)
980 # otherwise, s/?/shift @sqlv/ to put it inline
982 return ($sql, @sqlv);
984 1 while $sql =~ s/\?/my $d = shift(@sqlv);
985 ref $d ? $d->[1] : $d/e;
994 # This allows us to check for a local, then _form, attr
996 my($name) = $AUTOLOAD =~ /.*::(.+)/;
997 return $self->generate($name, @_);
1008 SQL::Abstract - Generate SQL from Perl data structures
1014 my $sql = SQL::Abstract->new;
1016 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1018 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1020 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1022 my($stmt, @bind) = $sql->delete($table, \%where);
1024 # Then, use these in your DBI statements
1025 my $sth = $dbh->prepare($stmt);
1026 $sth->execute(@bind);
1028 # Just generate the WHERE clause
1029 my($stmt, @bind) = $sql->where(\%where, \@order);
1031 # Return values in the same order, for hashed queries
1032 # See PERFORMANCE section for more details
1033 my @bind = $sql->values(\%fieldvals);
1037 This module was inspired by the excellent L<DBIx::Abstract>.
1038 However, in using that module I found that what I really wanted
1039 to do was generate SQL, but still retain complete control over my
1040 statement handles and use the DBI interface. So, I set out to
1041 create an abstract SQL generation module.
1043 While based on the concepts used by L<DBIx::Abstract>, there are
1044 several important differences, especially when it comes to WHERE
1045 clauses. I have modified the concepts used to make the SQL easier
1046 to generate from Perl data structures and, IMO, more intuitive.
1047 The underlying idea is for this module to do what you mean, based
1048 on the data structures you provide it. The big advantage is that
1049 you don't have to modify your code every time your data changes,
1050 as this module figures it out.
1052 To begin with, an SQL INSERT is as easy as just specifying a hash
1053 of C<key=value> pairs:
1056 name => 'Jimbo Bobson',
1057 phone => '123-456-7890',
1058 address => '42 Sister Lane',
1059 city => 'St. Louis',
1060 state => 'Louisiana',
1063 The SQL can then be generated with this:
1065 my($stmt, @bind) = $sql->insert('people', \%data);
1067 Which would give you something like this:
1069 $stmt = "INSERT INTO people
1070 (address, city, name, phone, state)
1071 VALUES (?, ?, ?, ?, ?)";
1072 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1073 '123-456-7890', 'Louisiana');
1075 These are then used directly in your DBI code:
1077 my $sth = $dbh->prepare($stmt);
1078 $sth->execute(@bind);
1080 =head2 Inserting and Updating Arrays
1082 If your database has array types (like for example Postgres),
1083 activate the special option C<< array_datatypes => 1 >>
1084 when creating the C<SQL::Abstract> object.
1085 Then you may use an arrayref to insert and update database array types:
1087 my $sql = SQL::Abstract->new(array_datatypes => 1);
1089 planets => [qw/Mercury Venus Earth Mars/]
1092 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1096 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1098 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1101 =head2 Inserting and Updating SQL
1103 In order to apply SQL functions to elements of your C<%data> you may
1104 specify a reference to an arrayref for the given hash value. For example,
1105 if you need to execute the Oracle C<to_date> function on a value, you can
1106 say something like this:
1110 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1113 The first value in the array is the actual SQL. Any other values are
1114 optional and would be included in the bind values array. This gives
1117 my($stmt, @bind) = $sql->insert('people', \%data);
1119 $stmt = "INSERT INTO people (name, date_entered)
1120 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1121 @bind = ('Bill', '03/02/2003');
1123 An UPDATE is just as easy, all you change is the name of the function:
1125 my($stmt, @bind) = $sql->update('people', \%data);
1127 Notice that your C<%data> isn't touched; the module will generate
1128 the appropriately quirky SQL for you automatically. Usually you'll
1129 want to specify a WHERE clause for your UPDATE, though, which is
1130 where handling C<%where> hashes comes in handy...
1132 =head2 Complex where statements
1134 This module can generate pretty complicated WHERE statements
1135 easily. For example, simple C<key=value> pairs are taken to mean
1136 equality, and if you want to see if a field is within a set
1137 of values, you can use an arrayref. Let's say we wanted to
1138 SELECT some data based on this criteria:
1141 requestor => 'inna',
1142 worker => ['nwiger', 'rcwe', 'sfz'],
1143 status => { '!=', 'completed' }
1146 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1148 The above would give you something like this:
1150 $stmt = "SELECT * FROM tickets WHERE
1151 ( requestor = ? ) AND ( status != ? )
1152 AND ( worker = ? OR worker = ? OR worker = ? )";
1153 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1155 Which you could then use in DBI code like so:
1157 my $sth = $dbh->prepare($stmt);
1158 $sth->execute(@bind);
1164 The functions are simple. There's one for each major SQL operation,
1165 and a constructor you use first. The arguments are specified in a
1166 similar order to each function (table, then fields, then a where
1167 clause) to try and simplify things.
1172 =head2 new(option => 'value')
1174 The C<new()> function takes a list of options and values, and returns
1175 a new B<SQL::Abstract> object which can then be used to generate SQL
1176 through the methods below. The options accepted are:
1182 If set to 'lower', then SQL will be generated in all lowercase. By
1183 default SQL is generated in "textbook" case meaning something like:
1185 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1187 Any setting other than 'lower' is ignored.
1191 This determines what the default comparison operator is. By default
1192 it is C<=>, meaning that a hash like this:
1194 %where = (name => 'nwiger', email => 'nate@wiger.org');
1196 Will generate SQL like this:
1198 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1200 However, you may want loose comparisons by default, so if you set
1201 C<cmp> to C<like> you would get SQL such as:
1203 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1205 You can also override the comparsion on an individual basis - see
1206 the huge section on L</"WHERE CLAUSES"> at the bottom.
1208 =item sqltrue, sqlfalse
1210 Expressions for inserting boolean values within SQL statements.
1211 By default these are C<1=1> and C<1=0>.
1215 This determines the default logical operator for multiple WHERE
1216 statements in arrays. By default it is "or", meaning that a WHERE
1220 event_date => {'>=', '2/13/99'},
1221 event_date => {'<=', '4/24/03'},
1224 Will generate SQL like this:
1226 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1228 This is probably not what you want given this query, though (look
1229 at the dates). To change the "OR" to an "AND", simply specify:
1231 my $sql = SQL::Abstract->new(logic => 'and');
1233 Which will change the above C<WHERE> to:
1235 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1237 The logic can also be changed locally by inserting
1238 an extra first element in the array :
1240 @where = (-and => event_date => {'>=', '2/13/99'},
1241 event_date => {'<=', '4/24/03'} );
1243 See the L</"WHERE CLAUSES"> section for explanations.
1247 This will automatically convert comparisons using the specified SQL
1248 function for both column and value. This is mostly used with an argument
1249 of C<upper> or C<lower>, so that the SQL will have the effect of
1250 case-insensitive "searches". For example, this:
1252 $sql = SQL::Abstract->new(convert => 'upper');
1253 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1255 Will turn out the following SQL:
1257 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1259 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1260 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1261 not validate this option; it will just pass through what you specify verbatim).
1265 This is a kludge because many databases suck. For example, you can't
1266 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1267 Instead, you have to use C<bind_param()>:
1269 $sth->bind_param(1, 'reg data');
1270 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1272 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1273 which loses track of which field each slot refers to. Fear not.
1275 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1276 Currently, you can specify either C<normal> (default) or C<columns>. If you
1277 specify C<columns>, you will get an array that looks like this:
1279 my $sql = SQL::Abstract->new(bindtype => 'columns');
1280 my($stmt, @bind) = $sql->insert(...);
1283 [ 'column1', 'value1' ],
1284 [ 'column2', 'value2' ],
1285 [ 'column3', 'value3' ],
1288 You can then iterate through this manually, using DBI's C<bind_param()>.
1290 $sth->prepare($stmt);
1293 my($col, $data) = @$_;
1294 if ($col eq 'details' || $col eq 'comments') {
1295 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1296 } elsif ($col eq 'image') {
1297 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1299 $sth->bind_param($i, $data);
1303 $sth->execute; # execute without @bind now
1305 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1306 Basically, the advantage is still that you don't have to care which fields
1307 are or are not included. You could wrap that above C<for> loop in a simple
1308 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1309 get a layer of abstraction over manual SQL specification.
1313 This is the character that a table or column name will be quoted
1314 with. By default this is an empty string, but you could set it to
1315 the character C<`>, to generate SQL like this:
1317 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1319 Alternatively, you can supply an array ref of two items, the first being the left
1320 hand quote character, and the second the right hand quote character. For
1321 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1322 that generates SQL like this:
1324 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1326 Quoting is useful if you have tables or columns names that are reserved
1327 words in your database's SQL dialect.
1331 This is the character that separates a table and column name. It is
1332 necessary to specify this when the C<quote_char> option is selected,
1333 so that tables and column names can be individually quoted like this:
1335 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1337 =item array_datatypes
1339 When this option is true, arrayrefs in INSERT or UPDATE are
1340 interpreted as array datatypes and are passed directly
1342 When this option is false, arrayrefs are interpreted
1343 as literal SQL, just like refs to arrayrefs
1344 (but this behavior is for backwards compatibility; when writing
1345 new queries, use the "reference to arrayref" syntax
1351 Takes a reference to a list of "special operators"
1352 to extend the syntax understood by L<SQL::Abstract>.
1353 See section L</"SPECIAL OPERATORS"> for details.
1359 =head2 insert($table, \@values || \%fieldvals)
1361 This is the simplest function. You simply give it a table name
1362 and either an arrayref of values or hashref of field/value pairs.
1363 It returns an SQL INSERT statement and a list of bind values.
1364 See the sections on L</"Inserting and Updating Arrays"> and
1365 L</"Inserting and Updating SQL"> for information on how to insert
1366 with those data types.
1368 =head2 update($table, \%fieldvals, \%where)
1370 This takes a table, hashref of field/value pairs, and an optional
1371 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1373 See the sections on L</"Inserting and Updating Arrays"> and
1374 L</"Inserting and Updating SQL"> for information on how to insert
1375 with those data types.
1377 =head2 select($source, $fields, $where, $order)
1379 This returns a SQL SELECT statement and associated list of bind values, as
1380 specified by the arguments :
1386 Specification of the 'FROM' part of the statement.
1387 The argument can be either a plain scalar (interpreted as a table
1388 name, will be quoted), or an arrayref (interpreted as a list
1389 of table names, joined by commas, quoted), or a scalarref
1390 (literal table name, not quoted), or a ref to an arrayref
1391 (list of literal table names, joined by commas, not quoted).
1395 Specification of the list of fields to retrieve from
1397 The argument can be either an arrayref (interpreted as a list
1398 of field names, will be joined by commas and quoted), or a
1399 plain scalar (literal SQL, not quoted).
1400 Please observe that this API is not as flexible as for
1401 the first argument C<$table>, for backwards compatibility reasons.
1405 Optional argument to specify the WHERE part of the query.
1406 The argument is most often a hashref, but can also be
1407 an arrayref or plain scalar --
1408 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
1412 Optional argument to specify the ORDER BY part of the query.
1413 The argument can be a scalar, a hashref or an arrayref
1414 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
1420 =head2 delete($table, \%where)
1422 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
1423 It returns an SQL DELETE statement and list of bind values.
1425 =head2 where(\%where, \@order)
1427 This is used to generate just the WHERE clause. For example,
1428 if you have an arbitrary data structure and know what the
1429 rest of your SQL is going to look like, but want an easy way
1430 to produce a WHERE clause, use this. It returns an SQL WHERE
1431 clause and list of bind values.
1434 =head2 values(\%data)
1436 This just returns the values from the hash C<%data>, in the same
1437 order that would be returned from any of the other above queries.
1438 Using this allows you to markedly speed up your queries if you
1439 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
1441 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
1443 Warning: This is an experimental method and subject to change.
1445 This returns arbitrarily generated SQL. It's a really basic shortcut.
1446 It will return two different things, depending on return context:
1448 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
1449 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
1451 These would return the following:
1453 # First calling form
1454 $stmt = "CREATE TABLE test (?, ?)";
1455 @bind = (field1, field2);
1457 # Second calling form
1458 $stmt_and_val = "CREATE TABLE test (field1, field2)";
1460 Depending on what you're trying to do, it's up to you to choose the correct
1461 format. In this example, the second form is what you would want.
1465 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
1469 ALTER SESSION SET nls_date_format = 'MM/YY'
1471 You get the idea. Strings get their case twiddled, but everything
1472 else remains verbatim.
1477 =head1 WHERE CLAUSES
1481 This module uses a variation on the idea from L<DBIx::Abstract>. It
1482 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
1483 module is that things in arrays are OR'ed, and things in hashes
1486 The easiest way to explain is to show lots of examples. After
1487 each C<%where> hash shown, it is assumed you used:
1489 my($stmt, @bind) = $sql->where(\%where);
1491 However, note that the C<%where> hash can be used directly in any
1492 of the other functions as well, as described above.
1494 =head2 Key-value pairs
1496 So, let's get started. To begin, a simple hash:
1500 status => 'completed'
1503 Is converted to SQL C<key = val> statements:
1505 $stmt = "WHERE user = ? AND status = ?";
1506 @bind = ('nwiger', 'completed');
1508 One common thing I end up doing is having a list of values that
1509 a field can be in. To do this, simply specify a list inside of
1514 status => ['assigned', 'in-progress', 'pending'];
1517 This simple code will create the following:
1519 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
1520 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
1522 An empty arrayref will be considered a logical false and
1525 =head2 Key-value pairs
1527 If you want to specify a different type of operator for your comparison,
1528 you can use a hashref for a given column:
1532 status => { '!=', 'completed' }
1535 Which would generate:
1537 $stmt = "WHERE user = ? AND status != ?";
1538 @bind = ('nwiger', 'completed');
1540 To test against multiple values, just enclose the values in an arrayref:
1542 status => { '!=', ['assigned', 'in-progress', 'pending'] };
1544 Which would give you:
1546 "WHERE status != ? AND status != ? AND status != ?"
1548 Notice that since the operator was recognized as being a 'negative'
1549 operator, the arrayref was interpreted with 'AND' logic (because
1550 of Morgan's laws). By contrast, the reverse
1552 status => { '=', ['assigned', 'in-progress', 'pending'] };
1556 "WHERE status = ? OR status = ? OR status = ?"
1559 The hashref can also contain multiple pairs, in which case it is expanded
1560 into an C<AND> of its elements:
1564 status => { '!=', 'completed', -not_like => 'pending%' }
1567 # Or more dynamically, like from a form
1568 $where{user} = 'nwiger';
1569 $where{status}{'!='} = 'completed';
1570 $where{status}{'-not_like'} = 'pending%';
1572 # Both generate this
1573 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
1574 @bind = ('nwiger', 'completed', 'pending%');
1577 To get an OR instead, you can combine it with the arrayref idea:
1581 priority => [ {'=', 2}, {'!=', 1} ]
1584 Which would generate:
1586 $stmt = "WHERE user = ? AND priority = ? OR priority != ?";
1587 @bind = ('nwiger', '2', '1');
1590 =head2 Logic and nesting operators
1592 In the example above,
1593 there is a subtle trap if you want to say something like
1594 this (notice the C<AND>):
1596 WHERE priority != ? AND priority != ?
1598 Because, in Perl you I<can't> do this:
1600 priority => { '!=', 2, '!=', 1 }
1602 As the second C<!=> key will obliterate the first. The solution
1603 is to use the special C<-modifier> form inside an arrayref:
1605 priority => [ -and => {'!=', 2},
1609 Normally, these would be joined by C<OR>, but the modifier tells it
1610 to use C<AND> instead. (Hint: You can use this in conjunction with the
1611 C<logic> option to C<new()> in order to change the way your queries
1612 work by default.) B<Important:> Note that the C<-modifier> goes
1613 B<INSIDE> the arrayref, as an extra first element. This will
1614 B<NOT> do what you think it might:
1616 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
1618 Here is a quick list of equivalencies, since there is some overlap:
1621 status => {'!=', 'completed', 'not like', 'pending%' }
1622 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
1625 status => {'=', ['assigned', 'in-progress']}
1626 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
1627 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
1629 In addition to C<-and> and C<-or>, there is also a special C<-nest>
1630 operator which adds an additional set of parens, to create a subquery.
1631 For example, to get something like this:
1633 $stmt = "WHERE user = ? AND ( workhrs > ? OR geo = ? )";
1634 @bind = ('nwiger', '20', 'ASIA');
1640 -nest => [ workhrs => {'>', 20}, geo => 'ASIA' ],
1643 If you need several nested subexpressions, you can number
1644 the C<-nest> branches :
1654 =head2 Special operators : IN, BETWEEN, etc.
1656 You can also use the hashref format to compare a list of fields using the
1657 C<IN> comparison operator, by specifying the list as an arrayref:
1660 status => 'completed',
1661 reportid => { -in => [567, 2335, 2] }
1664 Which would generate:
1666 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
1667 @bind = ('completed', '567', '2335', '2');
1669 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
1672 Another pair of operators is C<-between> and C<-not_between>,
1673 used with an arrayref of two values:
1677 completion_date => {
1678 -not_between => ['2002-10-01', '2003-02-06']
1684 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
1686 These are the two builtin "special operators"; but the
1687 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
1689 =head2 Nested conditions
1691 So far, we've seen how multiple conditions are joined with a top-level
1692 C<AND>. We can change this by putting the different conditions we want in
1693 hashes and then putting those hashes in an array. For example:
1698 status => { -like => ['pending%', 'dispatched'] },
1702 status => 'unassigned',
1706 This data structure would create the following:
1708 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
1709 OR ( user = ? AND status = ? ) )";
1710 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
1712 This can be combined with the C<-nest> operator to properly group
1719 ["-and", workhrs => {'>', 20}, geo => 'ASIA' ],
1720 ["-and", workhrs => {'<', 50}, geo => 'EURO' ]
1727 WHERE ( user = ? AND
1728 ( ( workhrs > ? AND geo = ? )
1729 OR ( workhrs < ? AND geo = ? ) ) )
1733 Finally, sometimes only literal SQL will do. If you want to include
1734 literal SQL verbatim, you can specify it as a scalar reference, namely:
1736 my $inn = 'is Not Null';
1738 priority => { '<', 2 },
1744 $stmt = "WHERE priority < ? AND requestor is Not Null";
1747 Note that in this example, you only get one bind parameter back, since
1748 the verbatim SQL is passed as part of the statement.
1750 Of course, just to prove a point, the above can also be accomplished
1754 priority => { '<', 2 },
1755 requestor => { '!=', undef },
1761 Conditions on boolean columns can be expressed in the
1762 same way, passing a reference to an empty string :
1765 priority => { '<', 2 },
1771 $stmt = "WHERE priority < ? AND is_ready";
1775 =head2 Literal SQL with placeholders and bind values (subqueries)
1777 If the literal SQL to be inserted has placeholders and bind values,
1778 use a reference to an arrayref (yes this is a double reference --
1779 not so common, but perfectly legal Perl). For example, to find a date
1780 in Postgres you can use something like this:
1783 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
1788 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
1792 Literal SQL is especially useful for nesting parenthesized clauses in the
1793 main SQL query. Here is a first example :
1795 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
1799 bar => \["IN ($sub_stmt)" => @sub_bind],
1804 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
1805 WHERE c2 < ? AND c3 LIKE ?))";
1806 @bind = (1234, 100, "foo%");
1808 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
1809 are expressed in the same way. Of course the C<$sub_stmt> and
1810 its associated bind values can be generated through a former call
1813 my ($sub_stmt, @sub_bind)
1814 = $sql->select("t1", "c1", {c2 => {"<" => 100},
1815 c3 => {-like => "foo%"}});
1818 bar => \["> ALL ($sub_stmt)" => @sub_bind],
1821 In the examples above, the subquery was used as an operator on a column;
1822 but the same principle also applies for a clause within the main C<%where>
1823 hash, like an EXISTS subquery :
1825 my ($sub_stmt, @sub_bind)
1826 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
1829 -nest => \["EXISTS ($sub_stmt)" => @sub_bind],
1834 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
1835 WHERE c1 = ? AND c2 > t0.c0))";
1839 Observe that the condition on C<c2> in the subquery refers to
1840 column C<t0.c0> of the main query : this is I<not> a bind
1841 value, so we have to express it through a scalar ref.
1842 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
1843 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
1844 what we wanted here.
1846 Another use of the subquery technique is when some SQL clauses need
1847 parentheses, as it often occurs with some proprietary SQL extensions
1848 like for example fulltext expressions, geospatial expressions,
1849 NATIVE clauses, etc. Here is an example of a fulltext query in MySQL :
1852 -nest => \["MATCH (col1, col2) AGAINST (?)" => qw/apples/]
1855 Finally, here is an example where a subquery is used
1856 for expressing unary negation:
1858 my ($sub_stmt, @sub_bind)
1859 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
1860 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
1862 lname => {like => '%son%'},
1863 -nest => \["NOT ($sub_stmt)" => @sub_bind],
1868 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
1869 @bind = ('%son%', 10, 20)
1875 These pages could go on for a while, since the nesting of the data
1876 structures this module can handle are pretty much unlimited (the
1877 module implements the C<WHERE> expansion as a recursive function
1878 internally). Your best bet is to "play around" with the module a
1879 little to see how the data structures behave, and choose the best
1880 format for your data based on that.
1882 And of course, all the values above will probably be replaced with
1883 variables gotten from forms or the command line. After all, if you
1884 knew everything ahead of time, you wouldn't have to worry about
1885 dynamically-generating SQL and could just hardwire it into your
1891 =head1 ORDER BY CLAUSES
1893 Some functions take an order by clause. This can either be a scalar (just a
1894 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
1895 or an array of either of the two previous forms. Examples:
1897 Given | Will Generate
1898 ----------------------------------------------------------
1899 \'colA DESC' | ORDER BY colA DESC
1900 'colA' | ORDER BY colA
1901 [qw/colA colB/] | ORDER BY colA, colB
1902 {-asc => 'colA'} | ORDER BY colA ASC
1903 {-desc => 'colB'} | ORDER BY colB DESC
1905 {-asc => 'colA'}, | ORDER BY colA ASC, colB DESC
1908 [colA => {-asc => 'colB'}] | ORDER BY colA, colB ASC
1909 ==========================================================
1913 =head1 SPECIAL OPERATORS
1915 my $sqlmaker = SQL::Abstract->new(special_ops => [
1918 my ($self, $field, $op, $arg) = @_;
1924 A "special operator" is a SQL syntactic clause that can be
1925 applied to a field, instead of a usual binary operator.
1928 WHERE field IN (?, ?, ?)
1929 WHERE field BETWEEN ? AND ?
1930 WHERE MATCH(field) AGAINST (?, ?)
1932 Special operators IN and BETWEEN are fairly standard and therefore
1933 are builtin within C<SQL::Abstract>. For other operators,
1934 like the MATCH .. AGAINST example above which is
1935 specific to MySQL, you can write your own operator handlers :
1936 supply a C<special_ops> argument to the C<new> method.
1937 That argument takes an arrayref of operator definitions;
1938 each operator definition is a hashref with two entries
1944 the regular expression to match the operator
1948 coderef that will be called when meeting that operator
1949 in the input tree. The coderef will be called with
1950 arguments C<< ($self, $field, $op, $arg) >>, and
1951 should return a C<< ($sql, @bind) >> structure.
1955 For example, here is an implementation
1956 of the MATCH .. AGAINST syntax for MySQL
1958 my $sqlmaker = SQL::Abstract->new(special_ops => [
1960 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
1961 {regex => qr/^match$/i,
1963 my ($self, $field, $op, $arg) = @_;
1964 $arg = [$arg] if not ref $arg;
1965 my $label = $self->_quote($field);
1966 my ($placeholder) = $self->_convert('?');
1967 my $placeholders = join ", ", (($placeholder) x @$arg);
1968 my $sql = $self->_sqlcase('match') . " ($label) "
1969 . $self->_sqlcase('against') . " ($placeholders) ";
1970 my @bind = $self->_bindtype($field, @$arg);
1971 return ($sql, @bind);
1980 Thanks to some benchmarking by Mark Stosberg, it turns out that
1981 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
1982 I must admit this wasn't an intentional design issue, but it's a
1983 byproduct of the fact that you get to control your C<DBI> handles
1986 To maximize performance, use a code snippet like the following:
1988 # prepare a statement handle using the first row
1989 # and then reuse it for the rest of the rows
1991 for my $href (@array_of_hashrefs) {
1992 $stmt ||= $sql->insert('table', $href);
1993 $sth ||= $dbh->prepare($stmt);
1994 $sth->execute($sql->values($href));
1997 The reason this works is because the keys in your C<$href> are sorted
1998 internally by B<SQL::Abstract>. Thus, as long as your data retains
1999 the same structure, you only have to generate the SQL the first time
2000 around. On subsequent queries, simply use the C<values> function provided
2001 by this module to return your values in the correct order.
2006 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2007 really like this part (I do, at least). Building up a complex query
2008 can be as simple as the following:
2012 use CGI::FormBuilder;
2015 my $form = CGI::FormBuilder->new(...);
2016 my $sql = SQL::Abstract->new;
2018 if ($form->submitted) {
2019 my $field = $form->field;
2020 my $id = delete $field->{id};
2021 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2024 Of course, you would still have to connect using C<DBI> to run the
2025 query, but the point is that if you make your form look like your
2026 table, the actual query script can be extremely simplistic.
2028 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2029 a fast interface to returning and formatting data. I frequently
2030 use these three modules together to write complex database query
2031 apps in under 50 lines.
2036 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2037 Great care has been taken to preserve the I<published> behavior
2038 documented in previous versions in the 1.* family; however,
2039 some features that were previously undocumented, or behaved
2040 differently from the documentation, had to be changed in order
2041 to clarify the semantics. Hence, client code that was relying
2042 on some dark areas of C<SQL::Abstract> v1.*
2043 B<might behave differently> in v1.50.
2045 The main changes are :
2051 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2055 added -nest1, -nest2 or -nest_1, -nest_2, ...
2059 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2063 defensive programming : check arguments
2067 fixed bug with global logic, which was previously implemented
2068 through global variables yielding side-effects. Prior versons would
2069 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2070 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2071 Now this is interpreted
2072 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2076 C<-and> / C<-or> operators are no longer accepted
2077 in the middle of an arrayref : they are
2078 only admitted if in first position.
2082 changed logic for distributing an op over arrayrefs
2086 fixed semantics of _bindtype on array args
2090 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2091 we just avoid shifting arrays within that tree.
2095 dropped the C<_modlogic> function
2101 =head1 ACKNOWLEDGEMENTS
2103 There are a number of individuals that have really helped out with
2104 this module. Unfortunately, most of them submitted bugs via CPAN
2105 so I have no idea who they are! But the people I do know are:
2107 Ash Berlin (order_by hash term support)
2108 Matt Trout (DBIx::Class support)
2109 Mark Stosberg (benchmarking)
2110 Chas Owens (initial "IN" operator support)
2111 Philip Collins (per-field SQL functions)
2112 Eric Kolve (hashref "AND" support)
2113 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2114 Dan Kubb (support for "quote_char" and "name_sep")
2115 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2116 Laurent Dami (internal refactoring, multiple -nest, extensible list of special operators, literal SQL)
2122 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
2126 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
2128 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
2130 For support, your best bet is to try the C<DBIx::Class> users mailing list.
2131 While not an official support venue, C<DBIx::Class> makes heavy use of
2132 C<SQL::Abstract>, and as such list members there are very familiar with
2133 how to create queries.
2135 This module is free software; you may copy this under the terms of
2136 the GNU General Public License, or the Artistic License, copies of
2137 which should have accompanied your Perl kit.