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)
13 use Data::Query::Constants qw(
14 DQ_IDENTIFIER DQ_OPERATOR DQ_VALUE DQ_LITERAL DQ_JOIN
17 #======================================================================
19 #======================================================================
21 our $VERSION = '1.72';
23 # This would confuse some packagers
24 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
28 # special operators (-in, -between). May be extended/overridden by user.
29 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
30 my @BUILTIN_SPECIAL_OPS = (
31 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
32 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
33 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
34 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
37 # unaryish operators - key maps to handler
38 my @BUILTIN_UNARY_OPS = (
39 # the digits are backcompat stuff
40 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
41 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
42 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
43 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
44 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
45 { regex => qr/^ value $/ix, handler => '_where_op_VALUE' },
48 #======================================================================
49 # DEBUGGING AND ERROR REPORTING
50 #======================================================================
53 return unless $_[0]->{debug}; shift; # a little faster
54 my $func = (caller(1))[3];
55 warn "[$func] ", @_, "\n";
59 my($func) = (caller(1))[3];
60 Carp::carp "[$func] Warning: ", @_;
64 my($func) = (caller(1))[3];
65 Carp::croak "[$func] Fatal: ", @_;
69 #======================================================================
71 #======================================================================
75 my $class = ref($self) || $self;
76 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
78 # choose our case by keeping an option around
79 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
81 # default logic for interpreting arrayrefs
82 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
84 # how to return bind vars
85 # LDNOTE: changed nwiger code : why this 'delete' ??
86 # $opt{bindtype} ||= delete($opt{bind_type}) || 'normal';
87 $opt{bindtype} ||= 'normal';
89 # default comparison is "=", but can be overridden
92 # try to recognize which are the 'equality' and 'unequality' ops
93 # (temporary quickfix, should go through a more seasoned API)
94 $opt{equality_op} = qr/^(\Q$opt{cmp}\E|is|(is\s+)?like)$/i;
95 $opt{inequality_op} = qr/^(!=|<>|(is\s+)?not(\s+like)?)$/i;
98 $opt{sqltrue} ||= '1=1';
99 $opt{sqlfalse} ||= '0=1';
102 $opt{special_ops} ||= [];
103 # regexes are applied in order, thus push after user-defines
104 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
107 $opt{unary_ops} ||= [];
108 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
110 # rudimentary saniy-check for user supplied bits treated as functions/operators
111 # If a purported function matches this regular expression, an exception is thrown.
112 # Literal SQL is *NOT* subject to this check, only functions (and column names
113 # when quoting is not in effect)
116 # need to guard against ()'s in column names too, but this will break tons of
117 # hacks... ideas anyone?
118 $opt{injection_guard} ||= qr/
124 $opt{name_sep} ||= '.';
126 $opt{renderer} ||= do {
127 require Data::Query::Renderer::SQL::Naive;
128 my ($always, $chars);
129 for ($opt{quote_char}) {
130 $chars = defined() ? (ref() ? $_ : [$_]) : ['',''];
133 Data::Query::Renderer::SQL::Naive->new({
134 quote_chars => $chars, always_quote => $always,
138 return bless \%opt, $class;
142 my ($self, $dq) = @_;
143 my ($sql, @bind) = @{$self->{renderer}->render($dq)};
144 wantarray ? ($sql, map $_->{value}, @bind) : $sql;
147 sub _assert_pass_injection_guard {
148 if ($_[1] =~ $_[0]->{injection_guard}) {
149 my $class = ref $_[0];
150 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
151 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
152 . "{injection_guard} attribute to ${class}->new()"
157 #======================================================================
159 #======================================================================
163 my $table = $self->_table(shift);
164 my $data = shift || return;
167 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
168 my ($sql, @bind) = $self->$method($data);
169 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
171 if ($options->{returning}) {
172 my ($s, @b) = $self->_insert_returning ($options);
177 return wantarray ? ($sql, @bind) : $sql;
180 sub _insert_returning {
181 my ($self, $options) = @_;
183 my $f = $options->{returning};
185 my $fieldlist = $self->_SWITCH_refkind($f, {
186 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
187 SCALAR => sub {$self->_quote($f)},
188 SCALARREF => sub {$$f},
190 return $self->_sqlcase(' returning ') . $fieldlist;
193 sub _insert_HASHREF { # explicit list of fields and then values
194 my ($self, $data) = @_;
196 my @fields = sort keys %$data;
198 my ($sql, @bind) = $self->_insert_values($data);
201 $_ = $self->_quote($_) foreach @fields;
202 $sql = "( ".join(", ", @fields).") ".$sql;
204 return ($sql, @bind);
207 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
208 my ($self, $data) = @_;
210 # no names (arrayref) so can't generate bindtype
211 $self->{bindtype} ne 'columns'
212 or belch "can't do 'columns' bindtype when called with arrayref";
214 # fold the list of values into a hash of column name - value pairs
215 # (where the column names are artificially generated, and their
216 # lexicographical ordering keep the ordering of the original list)
217 my $i = "a"; # incremented values will be in lexicographical order
218 my $data_in_hash = { map { ($i++ => $_) } @$data };
220 return $self->_insert_values($data_in_hash);
223 sub _insert_ARRAYREFREF { # literal SQL with bind
224 my ($self, $data) = @_;
226 my ($sql, @bind) = @${$data};
227 $self->_assert_bindval_matches_bindtype(@bind);
229 return ($sql, @bind);
233 sub _insert_SCALARREF { # literal SQL without bind
234 my ($self, $data) = @_;
240 my ($self, $data) = @_;
242 my (@values, @all_bind);
243 foreach my $column (sort keys %$data) {
244 my $v = $data->{$column};
246 $self->_SWITCH_refkind($v, {
249 if ($self->{array_datatypes}) { # if array datatype are activated
251 push @all_bind, $self->_bindtype($column, $v);
253 else { # else literal SQL with bind
254 my ($sql, @bind) = @$v;
255 $self->_assert_bindval_matches_bindtype(@bind);
257 push @all_bind, @bind;
261 ARRAYREFREF => sub { # literal SQL with bind
262 my ($sql, @bind) = @${$v};
263 $self->_assert_bindval_matches_bindtype(@bind);
265 push @all_bind, @bind;
268 # THINK : anything useful to do with a HASHREF ?
269 HASHREF => sub { # (nothing, but old SQLA passed it through)
270 #TODO in SQLA >= 2.0 it will die instead
271 belch "HASH ref as bind value in insert is not supported";
273 push @all_bind, $self->_bindtype($column, $v);
276 SCALARREF => sub { # literal SQL without bind
280 SCALAR_or_UNDEF => sub {
282 push @all_bind, $self->_bindtype($column, $v);
289 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
290 return ($sql, @all_bind);
295 #======================================================================
297 #======================================================================
302 my $table = $self->_table(shift);
303 my $data = shift || return;
306 # first build the 'SET' part of the sql statement
307 my (@set, @all_bind);
308 puke "Unsupported data type specified to \$sql->update"
309 unless ref $data eq 'HASH';
311 for my $k (sort keys %$data) {
314 my $label = $self->_quote($k);
316 $self->_SWITCH_refkind($v, {
318 if ($self->{array_datatypes}) { # array datatype
319 push @set, "$label = ?";
320 push @all_bind, $self->_bindtype($k, $v);
322 else { # literal SQL with bind
323 my ($sql, @bind) = @$v;
324 $self->_assert_bindval_matches_bindtype(@bind);
325 push @set, "$label = $sql";
326 push @all_bind, @bind;
329 ARRAYREFREF => sub { # literal SQL with bind
330 my ($sql, @bind) = @${$v};
331 $self->_assert_bindval_matches_bindtype(@bind);
332 push @set, "$label = $sql";
333 push @all_bind, @bind;
335 SCALARREF => sub { # literal SQL without bind
336 push @set, "$label = $$v";
339 my ($op, $arg, @rest) = %$v;
341 puke 'Operator calls in update must be in the form { -op => $arg }'
342 if (@rest or not $op =~ /^\-(.+)/);
344 local $self->{_nested_func_lhs} = $k;
345 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
347 push @set, "$label = $sql";
348 push @all_bind, @bind;
350 SCALAR_or_UNDEF => sub {
351 push @set, "$label = ?";
352 push @all_bind, $self->_bindtype($k, $v);
358 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
362 my($where_sql, @where_bind) = $self->where($where);
364 push @all_bind, @where_bind;
367 return wantarray ? ($sql, @all_bind) : $sql;
373 #======================================================================
375 #======================================================================
380 my $table = $self->_table(shift);
381 my $fields = shift || '*';
385 my($where_sql, @bind) = $self->where($where, $order);
387 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
389 my $sql = join(' ', $self->_sqlcase('select'), $f,
390 $self->_sqlcase('from'), $table)
393 return wantarray ? ($sql, @bind) : $sql;
396 #======================================================================
398 #======================================================================
403 my $table = $self->_table(shift);
407 my($where_sql, @bind) = $self->where($where);
408 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
410 return wantarray ? ($sql, @bind) : $sql;
414 #======================================================================
416 #======================================================================
420 # Finally, a separate routine just to handle WHERE clauses
422 my ($self, $where, $order) = @_;
425 my ($sql, @bind) = $self->_recurse_where($where);
426 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
430 $sql .= $self->_order_by($order);
433 return wantarray ? ($sql, @bind) : $sql;
438 my ($self, $where, $logic) = @_;
440 # dispatch on appropriate method according to refkind of $where
441 my $method = $self->_METHOD_FOR_refkind("_where", $where);
443 my ($sql, @bind) = $self->$method($where, $logic);
445 # DBIx::Class directly calls _recurse_where in scalar context, so
446 # we must implement it, even if not in the official API
447 return wantarray ? ($sql, @bind) : $sql;
452 #======================================================================
453 # WHERE: top-level ARRAYREF
454 #======================================================================
457 sub _where_ARRAYREF {
458 my ($self, $where, $logic) = @_;
460 $logic = uc($logic || $self->{logic});
461 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
463 my @clauses = @$where;
465 my (@sql_clauses, @all_bind);
466 # need to use while() so can shift() for pairs
467 while (my $el = shift @clauses) {
469 # switch according to kind of $el and get corresponding ($sql, @bind)
470 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
472 # skip empty elements, otherwise get invalid trailing AND stuff
473 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
477 $self->_assert_bindval_matches_bindtype(@b);
481 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
482 # LDNOTE : previous SQLA code for hashrefs was creating a dirty
483 # side-effect: the first hashref within an array would change
484 # the global logic to 'AND'. So [ {cond1, cond2}, [cond3, cond4] ]
485 # was interpreted as "(cond1 AND cond2) OR (cond3 AND cond4)",
486 # whereas it should be "(cond1 AND cond2) OR (cond3 OR cond4)".
488 SCALARREF => sub { ($$el); },
490 SCALAR => sub {# top-level arrayref with scalars, recurse in pairs
491 $self->_recurse_where({$el => shift(@clauses)})},
493 UNDEF => sub {puke "not supported : UNDEF in arrayref" },
497 push @sql_clauses, $sql;
498 push @all_bind, @bind;
502 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
505 #======================================================================
506 # WHERE: top-level ARRAYREFREF
507 #======================================================================
509 sub _where_ARRAYREFREF {
510 my ($self, $where) = @_;
511 my ($sql, @bind) = @$$where;
512 $self->_assert_bindval_matches_bindtype(@bind);
513 return ($sql, @bind);
516 #======================================================================
517 # WHERE: top-level HASHREF
518 #======================================================================
521 my ($self, $where) = @_;
522 my (@sql_clauses, @all_bind);
524 for my $k (sort keys %$where) {
525 my $v = $where->{$k};
527 # ($k => $v) is either a special unary op or a regular hashpair
528 my ($sql, @bind) = do {
530 # put the operator in canonical form
532 $op = substr $op, 1; # remove initial dash
533 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
534 $op =~ s/\s+/ /g; # compress whitespace
536 # so that -not_foo works correctly
537 $op =~ s/^not_/NOT /i;
539 $self->_debug("Unary OP(-$op) within hashref, recursing...");
540 my ($s, @b) = $self->_where_unary_op ($op, $v);
542 # top level vs nested
543 # we assume that handled unary ops will take care of their ()s
545 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
547 defined($self->{_nested_func_lhs}) && ($self->{_nested_func_lhs} eq $k)
552 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
553 $self->$method($k, $v);
557 push @sql_clauses, $sql;
558 push @all_bind, @bind;
561 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
564 sub _where_unary_op {
565 my ($self, $op, $rhs) = @_;
567 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
568 my $handler = $op_entry->{handler};
570 if (not ref $handler) {
571 if ($op =~ s/ [_\s]? \d+ $//x ) {
572 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
573 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
575 return $self->$handler ($op, $rhs);
577 elsif (ref $handler eq 'CODE') {
578 return $handler->($self, $op, $rhs);
581 puke "Illegal handler for operator $op - expecting a method name or a coderef";
585 $self->_debug("Generic unary OP: $op - recursing as function");
587 $self->_assert_pass_injection_guard($op);
589 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
591 puke "Illegal use of top-level '$op'"
592 unless $self->{_nested_func_lhs};
595 $self->_convert('?'),
596 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
600 $self->_recurse_where ($rhs)
604 $sql = sprintf ('%s %s',
605 $self->_sqlcase($op),
609 return ($sql, @bind);
612 sub _where_op_ANDOR {
613 my ($self, $op, $v) = @_;
615 $self->_SWITCH_refkind($v, {
617 return $self->_where_ARRAYREF($v, $op);
621 return ( $op =~ /^or/i )
622 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
623 : $self->_where_HASHREF($v);
627 puke "-$op => \\\$scalar makes little sense, use " .
629 ? '[ \$scalar, \%rest_of_conditions ] instead'
630 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
635 puke "-$op => \\[...] makes little sense, use " .
637 ? '[ \[...], \%rest_of_conditions ] instead'
638 : '-and => [ \[...], \%rest_of_conditions ] instead'
642 SCALAR => sub { # permissively interpreted as SQL
643 puke "-$op => \$value makes little sense, use -bool => \$value instead";
647 puke "-$op => undef not supported";
653 my ($self, $op, $v) = @_;
655 $self->_SWITCH_refkind($v, {
657 SCALAR => sub { # permissively interpreted as SQL
658 belch "literal SQL should be -nest => \\'scalar' "
659 . "instead of -nest => 'scalar' ";
664 puke "-$op => undef not supported";
668 $self->_recurse_where ($v);
676 my ($self, $op, $v) = @_;
678 my ($s, @b) = $self->_SWITCH_refkind($v, {
679 SCALAR => sub { # interpreted as SQL column
680 $self->_convert($self->_quote($v));
684 puke "-$op => undef not supported";
688 $self->_recurse_where ($v);
692 $s = "(NOT $s)" if $op =~ /^not/i;
697 sub _where_op_IDENT {
699 my ($op, $rhs) = splice @_, -2;
701 puke "-$op takes a single scalar argument (a quotable identifier)";
704 # in case we are called as a top level special op (no '=')
707 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
715 sub _where_op_VALUE {
717 my ($op, $rhs) = splice @_, -2;
719 # in case we are called as a top level special op (no '=')
724 ($lhs || $self->{_nested_func_lhs}),
731 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
735 $self->_convert('?'),
741 sub _where_hashpair_ARRAYREF {
742 my ($self, $k, $v) = @_;
745 my @v = @$v; # need copy because of shift below
746 $self->_debug("ARRAY($k) means distribute over elements");
748 # put apart first element if it is an operator (-and, -or)
750 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
754 my @distributed = map { {$k => $_} } @v;
757 $self->_debug("OP($op) reinjected into the distributed array");
758 unshift @distributed, $op;
761 my $logic = $op ? substr($op, 1) : '';
763 return $self->_recurse_where(\@distributed, $logic);
766 # LDNOTE : not sure of this one. What does "distribute over nothing" mean?
767 $self->_debug("empty ARRAY($k) means 0=1");
768 return ($self->{sqlfalse});
772 sub _where_hashpair_HASHREF {
773 my ($self, $k, $v, $logic) = @_;
776 local $self->{_nested_func_lhs} = $self->{_nested_func_lhs};
778 my ($all_sql, @all_bind);
780 for my $orig_op (sort keys %$v) {
781 my $val = $v->{$orig_op};
783 # put the operator in canonical form
786 # FIXME - we need to phase out dash-less ops
787 $op =~ s/^-//; # remove possible initial dash
788 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
789 $op =~ s/\s+/ /g; # compress whitespace
791 $self->_assert_pass_injection_guard($op);
793 # so that -not_foo works correctly
794 $op =~ s/^not_/NOT /i;
798 # CASE: col-value logic modifiers
799 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
800 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
802 # CASE: special operators like -in or -between
803 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
804 my $handler = $special_op->{handler};
806 puke "No handler supplied for special operator $orig_op";
808 elsif (not ref $handler) {
809 ($sql, @bind) = $self->$handler ($k, $op, $val);
811 elsif (ref $handler eq 'CODE') {
812 ($sql, @bind) = $handler->($self, $k, $op, $val);
815 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
819 $self->_SWITCH_refkind($val, {
821 ARRAYREF => sub { # CASE: col => {op => \@vals}
822 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
825 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
826 my ($sub_sql, @sub_bind) = @$$val;
827 $self->_assert_bindval_matches_bindtype(@sub_bind);
828 $sql = join ' ', $self->_convert($self->_quote($k)),
829 $self->_sqlcase($op),
834 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
835 my $is = ($op =~ $self->{equality_op}) ? 'is' :
836 ($op =~ $self->{inequality_op}) ? 'is not' :
837 puke "unexpected operator '$orig_op' with undef operand";
838 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
841 FALLBACK => sub { # CASE: col => {op/func => $stuff}
843 # retain for proper column type bind
844 $self->{_nested_func_lhs} ||= $k;
846 ($sql, @bind) = $self->_where_unary_op ($op, $val);
849 $self->_convert($self->_quote($k)),
850 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
856 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
857 push @all_bind, @bind;
859 return ($all_sql, @all_bind);
864 sub _where_field_op_ARRAYREF {
865 my ($self, $k, $op, $vals) = @_;
867 my @vals = @$vals; #always work on a copy
870 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
872 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
875 # see if the first element is an -and/-or op
877 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
882 # distribute $op over each remaining member of @vals, append logic if exists
883 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
885 # LDNOTE : had planned to change the distribution logic when
886 # $op =~ $self->{inequality_op}, because of Morgan laws :
887 # with {field => {'!=' => [22, 33]}}, it would be ridiculous to generate
888 # WHERE field != 22 OR field != 33 : the user probably means
889 # WHERE field != 22 AND field != 33.
890 # To do this, replace the above to roughly :
891 # my $logic = ($op =~ $self->{inequality_op}) ? 'AND' : 'OR';
892 # return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
896 # try to DWIM on equality operators
897 # LDNOTE : not 100% sure this is the correct thing to do ...
898 return ($self->{sqlfalse}) if $op =~ $self->{equality_op};
899 return ($self->{sqltrue}) if $op =~ $self->{inequality_op};
902 puke "operator '$op' applied on an empty array (field '$k')";
907 sub _where_hashpair_SCALARREF {
908 my ($self, $k, $v) = @_;
909 $self->_debug("SCALAR($k) means literal SQL: $$v");
910 my $sql = $self->_quote($k) . " " . $$v;
914 # literal SQL with bind
915 sub _where_hashpair_ARRAYREFREF {
916 my ($self, $k, $v) = @_;
917 $self->_debug("REF($k) means literal SQL: @${$v}");
918 my ($sql, @bind) = @$$v;
919 $self->_assert_bindval_matches_bindtype(@bind);
920 $sql = $self->_quote($k) . " " . $sql;
921 return ($sql, @bind );
924 # literal SQL without bind
925 sub _where_hashpair_SCALAR {
926 my ($self, $k, $v) = @_;
927 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
928 my $sql = join ' ', $self->_convert($self->_quote($k)),
929 $self->_sqlcase($self->{cmp}),
930 $self->_convert('?');
931 my @bind = $self->_bindtype($k, $v);
932 return ( $sql, @bind);
936 sub _where_hashpair_UNDEF {
937 my ($self, $k, $v) = @_;
938 $self->_debug("UNDEF($k) means IS NULL");
939 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
943 #======================================================================
944 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
945 #======================================================================
948 sub _where_SCALARREF {
949 my ($self, $where) = @_;
952 $self->_debug("SCALAR(*top) means literal SQL: $$where");
958 my ($self, $where) = @_;
961 $self->_debug("NOREF(*top) means literal SQL: $where");
972 #======================================================================
973 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
974 #======================================================================
977 sub _where_field_BETWEEN {
978 my ($self, $k, $op, $vals) = @_;
980 my ($label, $and, $placeholder);
981 $label = $self->_convert($self->_quote($k));
982 $and = ' ' . $self->_sqlcase('and') . ' ';
983 $placeholder = $self->_convert('?');
984 $op = $self->_sqlcase($op);
986 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
988 my ($s, @b) = @$$vals;
989 $self->_assert_bindval_matches_bindtype(@b);
996 puke "special op 'between' accepts an arrayref with exactly two values"
999 my (@all_sql, @all_bind);
1000 foreach my $val (@$vals) {
1001 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1003 return ($placeholder, $self->_bindtype($k, $val) );
1008 ARRAYREFREF => sub {
1009 my ($sql, @bind) = @$$val;
1010 $self->_assert_bindval_matches_bindtype(@bind);
1011 return ($sql, @bind);
1014 my ($func, $arg, @rest) = %$val;
1015 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1016 if (@rest or $func !~ /^ \- (.+)/x);
1017 local $self->{_nested_func_lhs} = $k;
1018 $self->_where_unary_op ($1 => $arg);
1021 push @all_sql, $sql;
1022 push @all_bind, @bind;
1026 (join $and, @all_sql),
1031 puke "special op 'between' accepts an arrayref with two values, or a single literal scalarref/arrayref-ref";
1035 my $sql = "( $label $op $clause )";
1036 return ($sql, @bind)
1040 sub _where_field_IN {
1041 my ($self, $k, $op, $vals) = @_;
1043 # backwards compatibility : if scalar, force into an arrayref
1044 $vals = [$vals] if defined $vals && ! ref $vals;
1046 my ($label) = $self->_convert($self->_quote($k));
1047 my ($placeholder) = $self->_convert('?');
1048 $op = $self->_sqlcase($op);
1050 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1051 ARRAYREF => sub { # list of choices
1052 if (@$vals) { # nonempty list
1053 my (@all_sql, @all_bind);
1055 for my $val (@$vals) {
1056 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1058 return ($placeholder, $val);
1063 ARRAYREFREF => sub {
1064 my ($sql, @bind) = @$$val;
1065 $self->_assert_bindval_matches_bindtype(@bind);
1066 return ($sql, @bind);
1069 my ($func, $arg, @rest) = %$val;
1070 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1071 if (@rest or $func !~ /^ \- (.+)/x);
1072 local $self->{_nested_func_lhs} = $k;
1073 $self->_where_unary_op ($1 => $arg);
1076 return $self->_sqlcase('null');
1079 push @all_sql, $sql;
1080 push @all_bind, @bind;
1084 sprintf ('%s %s ( %s )',
1087 join (', ', @all_sql)
1089 $self->_bindtype($k, @all_bind),
1092 else { # empty list : some databases won't understand "IN ()", so DWIM
1093 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1098 SCALARREF => sub { # literal SQL
1099 my $sql = $self->_open_outer_paren ($$vals);
1100 return ("$label $op ( $sql )");
1102 ARRAYREFREF => sub { # literal SQL with bind
1103 my ($sql, @bind) = @$$vals;
1104 $self->_assert_bindval_matches_bindtype(@bind);
1105 $sql = $self->_open_outer_paren ($sql);
1106 return ("$label $op ( $sql )", @bind);
1110 puke "special op 'in' requires an arrayref (or scalarref/arrayref-ref)";
1114 return ($sql, @bind);
1117 # Some databases (SQLite) treat col IN (1, 2) different from
1118 # col IN ( (1, 2) ). Use this to strip all outer parens while
1119 # adding them back in the corresponding method
1120 sub _open_outer_paren {
1121 my ($self, $sql) = @_;
1122 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/xs;
1127 #======================================================================
1129 #======================================================================
1132 my ($self, $arg) = @_;
1135 for my $c ($self->_order_by_chunks ($arg) ) {
1136 $self->_SWITCH_refkind ($c, {
1137 SCALAR => sub { push @sql, $c },
1138 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1144 $self->_sqlcase(' order by'),
1150 return wantarray ? ($sql, @bind) : $sql;
1153 sub _order_by_chunks {
1154 my ($self, $arg) = @_;
1156 return $self->_SWITCH_refkind($arg, {
1159 map { $self->_order_by_chunks ($_ ) } @$arg;
1162 ARRAYREFREF => sub {
1163 my ($s, @b) = @$$arg;
1164 $self->_assert_bindval_matches_bindtype(@b);
1168 SCALAR => sub {$self->_quote($arg)},
1170 UNDEF => sub {return () },
1172 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1175 # get first pair in hash
1176 my ($key, $val, @rest) = %$arg;
1178 return () unless $key;
1180 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1181 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1187 for my $c ($self->_order_by_chunks ($val)) {
1190 $self->_SWITCH_refkind ($c, {
1195 ($sql, @bind) = @$c;
1199 $sql = $sql . ' ' . $self->_sqlcase($direction);
1201 push @ret, [ $sql, @bind];
1210 #======================================================================
1211 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1212 #======================================================================
1218 $self->_SWITCH_refkind($from, {
1220 die "Empty FROM list" unless my @f = @$from;
1222 type => DQ_IDENTIFIER,
1223 elements => [ split /\Q$self->{name_sep}/, shift @f ],
1225 while (my $x = shift @f) {
1229 type => DQ_IDENTIFIER,
1230 elements => [ split /\Q$self->{name_sep}/, $x ],
1238 type => DQ_IDENTIFIER,
1239 elements => [ split /\Q$self->{name_sep}/, $from ],
1254 #======================================================================
1256 #======================================================================
1258 # highly optimized, as it's called way too often
1260 # my ($self, $label) = @_;
1262 return '' unless defined $_[1];
1263 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1265 unless ($_[0]->{quote_char}) {
1266 $_[0]->_assert_pass_injection_guard($_[1]);
1270 my $qref = ref $_[0]->{quote_char};
1273 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1275 elsif ($qref eq 'ARRAY') {
1276 ($l, $r) = @{$_[0]->{quote_char}};
1279 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1282 # parts containing * are naturally unquoted
1283 return join( $_[0]->{name_sep}||'', map
1284 { $_ eq '*' ? $_ : $l . $_ . $r }
1285 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1290 # Conversion, if applicable
1292 #my ($self, $arg) = @_;
1294 # LDNOTE : modified the previous implementation below because
1295 # it was not consistent : the first "return" is always an array,
1296 # the second "return" is context-dependent. Anyway, _convert
1297 # seems always used with just a single argument, so make it a
1299 # return @_ unless $self->{convert};
1300 # my $conv = $self->_sqlcase($self->{convert});
1301 # my @ret = map { $conv.'('.$_.')' } @_;
1302 # return wantarray ? @ret : $ret[0];
1303 if ($_[0]->{convert}) {
1304 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1311 #my ($self, $col, @vals) = @_;
1313 #LDNOTE : changed original implementation below because it did not make
1314 # sense when bindtype eq 'columns' and @vals > 1.
1315 # return $self->{bindtype} eq 'columns' ? [ $col, @vals ] : @vals;
1317 # called often - tighten code
1318 return $_[0]->{bindtype} eq 'columns'
1319 ? map {[$_[1], $_]} @_[2 .. $#_]
1324 # Dies if any element of @bind is not in [colname => value] format
1325 # if bindtype is 'columns'.
1326 sub _assert_bindval_matches_bindtype {
1327 # my ($self, @bind) = @_;
1329 if ($self->{bindtype} eq 'columns') {
1331 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1332 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1338 sub _join_sql_clauses {
1339 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1341 if (@$clauses_aref > 1) {
1342 my $join = " " . $self->_sqlcase($logic) . " ";
1343 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1344 return ($sql, @$bind_aref);
1346 elsif (@$clauses_aref) {
1347 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1350 return (); # if no SQL, ignore @$bind_aref
1355 # Fix SQL case, if so requested
1357 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1358 # don't touch the argument ... crooked logic, but let's not change it!
1359 return $_[0]->{case} ? $_[1] : uc($_[1]);
1363 #======================================================================
1364 # DISPATCHING FROM REFKIND
1365 #======================================================================
1368 my ($self, $data) = @_;
1370 return 'UNDEF' unless defined $data;
1372 # blessed objects are treated like scalars
1373 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1375 return 'SCALAR' unless $ref;
1378 while ($ref eq 'REF') {
1380 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1384 return ($ref||'SCALAR') . ('REF' x $n_steps);
1388 my ($self, $data) = @_;
1389 my @try = ($self->_refkind($data));
1390 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1391 push @try, 'FALLBACK';
1395 sub _METHOD_FOR_refkind {
1396 my ($self, $meth_prefix, $data) = @_;
1399 for (@{$self->_try_refkind($data)}) {
1400 $method = $self->can($meth_prefix."_".$_)
1404 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1408 sub _SWITCH_refkind {
1409 my ($self, $data, $dispatch_table) = @_;
1412 for (@{$self->_try_refkind($data)}) {
1413 $coderef = $dispatch_table->{$_}
1417 puke "no dispatch entry for ".$self->_refkind($data)
1426 #======================================================================
1427 # VALUES, GENERATE, AUTOLOAD
1428 #======================================================================
1430 # LDNOTE: original code from nwiger, didn't touch code in that section
1431 # I feel the AUTOLOAD stuff should not be the default, it should
1432 # only be activated on explicit demand by user.
1436 my $data = shift || return;
1437 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1438 unless ref $data eq 'HASH';
1441 foreach my $k ( sort keys %$data ) {
1442 my $v = $data->{$k};
1443 $self->_SWITCH_refkind($v, {
1445 if ($self->{array_datatypes}) { # array datatype
1446 push @all_bind, $self->_bindtype($k, $v);
1448 else { # literal SQL with bind
1449 my ($sql, @bind) = @$v;
1450 $self->_assert_bindval_matches_bindtype(@bind);
1451 push @all_bind, @bind;
1454 ARRAYREFREF => sub { # literal SQL with bind
1455 my ($sql, @bind) = @${$v};
1456 $self->_assert_bindval_matches_bindtype(@bind);
1457 push @all_bind, @bind;
1459 SCALARREF => sub { # literal SQL without bind
1461 SCALAR_or_UNDEF => sub {
1462 push @all_bind, $self->_bindtype($k, $v);
1473 my(@sql, @sqlq, @sqlv);
1477 if ($ref eq 'HASH') {
1478 for my $k (sort keys %$_) {
1481 my $label = $self->_quote($k);
1482 if ($r eq 'ARRAY') {
1483 # literal SQL with bind
1484 my ($sql, @bind) = @$v;
1485 $self->_assert_bindval_matches_bindtype(@bind);
1486 push @sqlq, "$label = $sql";
1488 } elsif ($r eq 'SCALAR') {
1489 # literal SQL without bind
1490 push @sqlq, "$label = $$v";
1492 push @sqlq, "$label = ?";
1493 push @sqlv, $self->_bindtype($k, $v);
1496 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1497 } elsif ($ref eq 'ARRAY') {
1498 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1501 if ($r eq 'ARRAY') { # literal SQL with bind
1502 my ($sql, @bind) = @$v;
1503 $self->_assert_bindval_matches_bindtype(@bind);
1506 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1507 # embedded literal SQL
1514 push @sql, '(' . join(', ', @sqlq) . ')';
1515 } elsif ($ref eq 'SCALAR') {
1519 # strings get case twiddled
1520 push @sql, $self->_sqlcase($_);
1524 my $sql = join ' ', @sql;
1526 # this is pretty tricky
1527 # if ask for an array, return ($stmt, @bind)
1528 # otherwise, s/?/shift @sqlv/ to put it inline
1530 return ($sql, @sqlv);
1532 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1533 ref $d ? $d->[1] : $d/e;
1542 # This allows us to check for a local, then _form, attr
1544 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1545 return $self->generate($name, @_);
1556 SQL::Abstract - Generate SQL from Perl data structures
1562 my $sql = SQL::Abstract->new;
1564 my($stmt, @bind) = $sql->select($table, \@fields, \%where, \@order);
1566 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1568 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1570 my($stmt, @bind) = $sql->delete($table, \%where);
1572 # Then, use these in your DBI statements
1573 my $sth = $dbh->prepare($stmt);
1574 $sth->execute(@bind);
1576 # Just generate the WHERE clause
1577 my($stmt, @bind) = $sql->where(\%where, \@order);
1579 # Return values in the same order, for hashed queries
1580 # See PERFORMANCE section for more details
1581 my @bind = $sql->values(\%fieldvals);
1585 This module was inspired by the excellent L<DBIx::Abstract>.
1586 However, in using that module I found that what I really wanted
1587 to do was generate SQL, but still retain complete control over my
1588 statement handles and use the DBI interface. So, I set out to
1589 create an abstract SQL generation module.
1591 While based on the concepts used by L<DBIx::Abstract>, there are
1592 several important differences, especially when it comes to WHERE
1593 clauses. I have modified the concepts used to make the SQL easier
1594 to generate from Perl data structures and, IMO, more intuitive.
1595 The underlying idea is for this module to do what you mean, based
1596 on the data structures you provide it. The big advantage is that
1597 you don't have to modify your code every time your data changes,
1598 as this module figures it out.
1600 To begin with, an SQL INSERT is as easy as just specifying a hash
1601 of C<key=value> pairs:
1604 name => 'Jimbo Bobson',
1605 phone => '123-456-7890',
1606 address => '42 Sister Lane',
1607 city => 'St. Louis',
1608 state => 'Louisiana',
1611 The SQL can then be generated with this:
1613 my($stmt, @bind) = $sql->insert('people', \%data);
1615 Which would give you something like this:
1617 $stmt = "INSERT INTO people
1618 (address, city, name, phone, state)
1619 VALUES (?, ?, ?, ?, ?)";
1620 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1621 '123-456-7890', 'Louisiana');
1623 These are then used directly in your DBI code:
1625 my $sth = $dbh->prepare($stmt);
1626 $sth->execute(@bind);
1628 =head2 Inserting and Updating Arrays
1630 If your database has array types (like for example Postgres),
1631 activate the special option C<< array_datatypes => 1 >>
1632 when creating the C<SQL::Abstract> object.
1633 Then you may use an arrayref to insert and update database array types:
1635 my $sql = SQL::Abstract->new(array_datatypes => 1);
1637 planets => [qw/Mercury Venus Earth Mars/]
1640 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1644 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1646 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1649 =head2 Inserting and Updating SQL
1651 In order to apply SQL functions to elements of your C<%data> you may
1652 specify a reference to an arrayref for the given hash value. For example,
1653 if you need to execute the Oracle C<to_date> function on a value, you can
1654 say something like this:
1658 date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
1661 The first value in the array is the actual SQL. Any other values are
1662 optional and would be included in the bind values array. This gives
1665 my($stmt, @bind) = $sql->insert('people', \%data);
1667 $stmt = "INSERT INTO people (name, date_entered)
1668 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1669 @bind = ('Bill', '03/02/2003');
1671 An UPDATE is just as easy, all you change is the name of the function:
1673 my($stmt, @bind) = $sql->update('people', \%data);
1675 Notice that your C<%data> isn't touched; the module will generate
1676 the appropriately quirky SQL for you automatically. Usually you'll
1677 want to specify a WHERE clause for your UPDATE, though, which is
1678 where handling C<%where> hashes comes in handy...
1680 =head2 Complex where statements
1682 This module can generate pretty complicated WHERE statements
1683 easily. For example, simple C<key=value> pairs are taken to mean
1684 equality, and if you want to see if a field is within a set
1685 of values, you can use an arrayref. Let's say we wanted to
1686 SELECT some data based on this criteria:
1689 requestor => 'inna',
1690 worker => ['nwiger', 'rcwe', 'sfz'],
1691 status => { '!=', 'completed' }
1694 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1696 The above would give you something like this:
1698 $stmt = "SELECT * FROM tickets WHERE
1699 ( requestor = ? ) AND ( status != ? )
1700 AND ( worker = ? OR worker = ? OR worker = ? )";
1701 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1703 Which you could then use in DBI code like so:
1705 my $sth = $dbh->prepare($stmt);
1706 $sth->execute(@bind);
1712 The functions are simple. There's one for each major SQL operation,
1713 and a constructor you use first. The arguments are specified in a
1714 similar order to each function (table, then fields, then a where
1715 clause) to try and simplify things.
1720 =head2 new(option => 'value')
1722 The C<new()> function takes a list of options and values, and returns
1723 a new B<SQL::Abstract> object which can then be used to generate SQL
1724 through the methods below. The options accepted are:
1730 If set to 'lower', then SQL will be generated in all lowercase. By
1731 default SQL is generated in "textbook" case meaning something like:
1733 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1735 Any setting other than 'lower' is ignored.
1739 This determines what the default comparison operator is. By default
1740 it is C<=>, meaning that a hash like this:
1742 %where = (name => 'nwiger', email => 'nate@wiger.org');
1744 Will generate SQL like this:
1746 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1748 However, you may want loose comparisons by default, so if you set
1749 C<cmp> to C<like> you would get SQL such as:
1751 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1753 You can also override the comparsion on an individual basis - see
1754 the huge section on L</"WHERE CLAUSES"> at the bottom.
1756 =item sqltrue, sqlfalse
1758 Expressions for inserting boolean values within SQL statements.
1759 By default these are C<1=1> and C<1=0>. They are used
1760 by the special operators C<-in> and C<-not_in> for generating
1761 correct SQL even when the argument is an empty array (see below).
1765 This determines the default logical operator for multiple WHERE
1766 statements in arrays or hashes. If absent, the default logic is "or"
1767 for arrays, and "and" for hashes. This means that a WHERE
1771 event_date => {'>=', '2/13/99'},
1772 event_date => {'<=', '4/24/03'},
1775 will generate SQL like this:
1777 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1779 This is probably not what you want given this query, though (look
1780 at the dates). To change the "OR" to an "AND", simply specify:
1782 my $sql = SQL::Abstract->new(logic => 'and');
1784 Which will change the above C<WHERE> to:
1786 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1788 The logic can also be changed locally by inserting
1789 a modifier in front of an arrayref :
1791 @where = (-and => [event_date => {'>=', '2/13/99'},
1792 event_date => {'<=', '4/24/03'} ]);
1794 See the L</"WHERE CLAUSES"> section for explanations.
1798 This will automatically convert comparisons using the specified SQL
1799 function for both column and value. This is mostly used with an argument
1800 of C<upper> or C<lower>, so that the SQL will have the effect of
1801 case-insensitive "searches". For example, this:
1803 $sql = SQL::Abstract->new(convert => 'upper');
1804 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1806 Will turn out the following SQL:
1808 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1810 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1811 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1812 not validate this option; it will just pass through what you specify verbatim).
1816 This is a kludge because many databases suck. For example, you can't
1817 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1818 Instead, you have to use C<bind_param()>:
1820 $sth->bind_param(1, 'reg data');
1821 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1823 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1824 which loses track of which field each slot refers to. Fear not.
1826 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1827 Currently, you can specify either C<normal> (default) or C<columns>. If you
1828 specify C<columns>, you will get an array that looks like this:
1830 my $sql = SQL::Abstract->new(bindtype => 'columns');
1831 my($stmt, @bind) = $sql->insert(...);
1834 [ 'column1', 'value1' ],
1835 [ 'column2', 'value2' ],
1836 [ 'column3', 'value3' ],
1839 You can then iterate through this manually, using DBI's C<bind_param()>.
1841 $sth->prepare($stmt);
1844 my($col, $data) = @$_;
1845 if ($col eq 'details' || $col eq 'comments') {
1846 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1847 } elsif ($col eq 'image') {
1848 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1850 $sth->bind_param($i, $data);
1854 $sth->execute; # execute without @bind now
1856 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1857 Basically, the advantage is still that you don't have to care which fields
1858 are or are not included. You could wrap that above C<for> loop in a simple
1859 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1860 get a layer of abstraction over manual SQL specification.
1862 Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
1863 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1864 will expect the bind values in this format.
1868 This is the character that a table or column name will be quoted
1869 with. By default this is an empty string, but you could set it to
1870 the character C<`>, to generate SQL like this:
1872 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1874 Alternatively, you can supply an array ref of two items, the first being the left
1875 hand quote character, and the second the right hand quote character. For
1876 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1877 that generates SQL like this:
1879 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1881 Quoting is useful if you have tables or columns names that are reserved
1882 words in your database's SQL dialect.
1886 This is the character that separates a table and column name. It is
1887 necessary to specify this when the C<quote_char> option is selected,
1888 so that tables and column names can be individually quoted like this:
1890 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1892 =item injection_guard
1894 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1895 column name specified in a query structure. This is a safety mechanism to avoid
1896 injection attacks when mishandling user input e.g.:
1898 my %condition_as_column_value_pairs = get_values_from_user();
1899 $sqla->select( ... , \%condition_as_column_value_pairs );
1901 If the expression matches an exception is thrown. Note that literal SQL
1902 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1904 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1906 =item array_datatypes
1908 When this option is true, arrayrefs in INSERT or UPDATE are
1909 interpreted as array datatypes and are passed directly
1911 When this option is false, arrayrefs are interpreted
1912 as literal SQL, just like refs to arrayrefs
1913 (but this behavior is for backwards compatibility; when writing
1914 new queries, use the "reference to arrayref" syntax
1920 Takes a reference to a list of "special operators"
1921 to extend the syntax understood by L<SQL::Abstract>.
1922 See section L</"SPECIAL OPERATORS"> for details.
1926 Takes a reference to a list of "unary operators"
1927 to extend the syntax understood by L<SQL::Abstract>.
1928 See section L</"UNARY OPERATORS"> for details.
1934 =head2 insert($table, \@values || \%fieldvals, \%options)
1936 This is the simplest function. You simply give it a table name
1937 and either an arrayref of values or hashref of field/value pairs.
1938 It returns an SQL INSERT statement and a list of bind values.
1939 See the sections on L</"Inserting and Updating Arrays"> and
1940 L</"Inserting and Updating SQL"> for information on how to insert
1941 with those data types.
1943 The optional C<\%options> hash reference may contain additional
1944 options to generate the insert SQL. Currently supported options
1951 Takes either a scalar of raw SQL fields, or an array reference of
1952 field names, and adds on an SQL C<RETURNING> statement at the end.
1953 This allows you to return data generated by the insert statement
1954 (such as row IDs) without performing another C<SELECT> statement.
1955 Note, however, this is not part of the SQL standard and may not
1956 be supported by all database engines.
1960 =head2 update($table, \%fieldvals, \%where)
1962 This takes a table, hashref of field/value pairs, and an optional
1963 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
1965 See the sections on L</"Inserting and Updating Arrays"> and
1966 L</"Inserting and Updating SQL"> for information on how to insert
1967 with those data types.
1969 =head2 select($source, $fields, $where, $order)
1971 This returns a SQL SELECT statement and associated list of bind values, as
1972 specified by the arguments :
1978 Specification of the 'FROM' part of the statement.
1979 The argument can be either a plain scalar (interpreted as a table
1980 name, will be quoted), or an arrayref (interpreted as a list
1981 of table names, joined by commas, quoted), or a scalarref
1982 (literal table name, not quoted), or a ref to an arrayref
1983 (list of literal table names, joined by commas, not quoted).
1987 Specification of the list of fields to retrieve from
1989 The argument can be either an arrayref (interpreted as a list
1990 of field names, will be joined by commas and quoted), or a
1991 plain scalar (literal SQL, not quoted).
1992 Please observe that this API is not as flexible as for
1993 the first argument C<$table>, for backwards compatibility reasons.
1997 Optional argument to specify the WHERE part of the query.
1998 The argument is most often a hashref, but can also be
1999 an arrayref or plain scalar --
2000 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2004 Optional argument to specify the ORDER BY part of the query.
2005 The argument can be a scalar, a hashref or an arrayref
2006 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2012 =head2 delete($table, \%where)
2014 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2015 It returns an SQL DELETE statement and list of bind values.
2017 =head2 where(\%where, \@order)
2019 This is used to generate just the WHERE clause. For example,
2020 if you have an arbitrary data structure and know what the
2021 rest of your SQL is going to look like, but want an easy way
2022 to produce a WHERE clause, use this. It returns an SQL WHERE
2023 clause and list of bind values.
2026 =head2 values(\%data)
2028 This just returns the values from the hash C<%data>, in the same
2029 order that would be returned from any of the other above queries.
2030 Using this allows you to markedly speed up your queries if you
2031 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2033 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2035 Warning: This is an experimental method and subject to change.
2037 This returns arbitrarily generated SQL. It's a really basic shortcut.
2038 It will return two different things, depending on return context:
2040 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2041 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2043 These would return the following:
2045 # First calling form
2046 $stmt = "CREATE TABLE test (?, ?)";
2047 @bind = (field1, field2);
2049 # Second calling form
2050 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2052 Depending on what you're trying to do, it's up to you to choose the correct
2053 format. In this example, the second form is what you would want.
2057 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2061 ALTER SESSION SET nls_date_format = 'MM/YY'
2063 You get the idea. Strings get their case twiddled, but everything
2064 else remains verbatim.
2066 =head1 WHERE CLAUSES
2070 This module uses a variation on the idea from L<DBIx::Abstract>. It
2071 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2072 module is that things in arrays are OR'ed, and things in hashes
2075 The easiest way to explain is to show lots of examples. After
2076 each C<%where> hash shown, it is assumed you used:
2078 my($stmt, @bind) = $sql->where(\%where);
2080 However, note that the C<%where> hash can be used directly in any
2081 of the other functions as well, as described above.
2083 =head2 Key-value pairs
2085 So, let's get started. To begin, a simple hash:
2089 status => 'completed'
2092 Is converted to SQL C<key = val> statements:
2094 $stmt = "WHERE user = ? AND status = ?";
2095 @bind = ('nwiger', 'completed');
2097 One common thing I end up doing is having a list of values that
2098 a field can be in. To do this, simply specify a list inside of
2103 status => ['assigned', 'in-progress', 'pending'];
2106 This simple code will create the following:
2108 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2109 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2111 A field associated to an empty arrayref will be considered a
2112 logical false and will generate 0=1.
2114 =head2 Tests for NULL values
2116 If the value part is C<undef> then this is converted to SQL <IS NULL>
2125 $stmt = "WHERE user = ? AND status IS NULL";
2128 To test if a column IS NOT NULL:
2132 status => { '!=', undef },
2135 =head2 Specific comparison operators
2137 If you want to specify a different type of operator for your comparison,
2138 you can use a hashref for a given column:
2142 status => { '!=', 'completed' }
2145 Which would generate:
2147 $stmt = "WHERE user = ? AND status != ?";
2148 @bind = ('nwiger', 'completed');
2150 To test against multiple values, just enclose the values in an arrayref:
2152 status => { '=', ['assigned', 'in-progress', 'pending'] };
2154 Which would give you:
2156 "WHERE status = ? OR status = ? OR status = ?"
2159 The hashref can also contain multiple pairs, in which case it is expanded
2160 into an C<AND> of its elements:
2164 status => { '!=', 'completed', -not_like => 'pending%' }
2167 # Or more dynamically, like from a form
2168 $where{user} = 'nwiger';
2169 $where{status}{'!='} = 'completed';
2170 $where{status}{'-not_like'} = 'pending%';
2172 # Both generate this
2173 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2174 @bind = ('nwiger', 'completed', 'pending%');
2177 To get an OR instead, you can combine it with the arrayref idea:
2181 priority => [ { '=', 2 }, { '>', 5 } ]
2184 Which would generate:
2186 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2187 @bind = ('2', '5', 'nwiger');
2189 If you want to include literal SQL (with or without bind values), just use a
2190 scalar reference or array reference as the value:
2193 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2194 date_expires => { '<' => \"now()" }
2197 Which would generate:
2199 $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2200 @bind = ('11/26/2008');
2203 =head2 Logic and nesting operators
2205 In the example above,
2206 there is a subtle trap if you want to say something like
2207 this (notice the C<AND>):
2209 WHERE priority != ? AND priority != ?
2211 Because, in Perl you I<can't> do this:
2213 priority => { '!=', 2, '!=', 1 }
2215 As the second C<!=> key will obliterate the first. The solution
2216 is to use the special C<-modifier> form inside an arrayref:
2218 priority => [ -and => {'!=', 2},
2222 Normally, these would be joined by C<OR>, but the modifier tells it
2223 to use C<AND> instead. (Hint: You can use this in conjunction with the
2224 C<logic> option to C<new()> in order to change the way your queries
2225 work by default.) B<Important:> Note that the C<-modifier> goes
2226 B<INSIDE> the arrayref, as an extra first element. This will
2227 B<NOT> do what you think it might:
2229 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2231 Here is a quick list of equivalencies, since there is some overlap:
2234 status => {'!=', 'completed', 'not like', 'pending%' }
2235 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2238 status => {'=', ['assigned', 'in-progress']}
2239 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2240 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2244 =head2 Special operators : IN, BETWEEN, etc.
2246 You can also use the hashref format to compare a list of fields using the
2247 C<IN> comparison operator, by specifying the list as an arrayref:
2250 status => 'completed',
2251 reportid => { -in => [567, 2335, 2] }
2254 Which would generate:
2256 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2257 @bind = ('completed', '567', '2335', '2');
2259 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2262 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2263 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2264 'sqltrue' (by default : C<1=1>).
2266 In addition to the array you can supply a chunk of literal sql or
2267 literal sql with bind:
2270 customer => { -in => \[
2271 'SELECT cust_id FROM cust WHERE balance > ?',
2274 status => { -in => \'SELECT status_codes FROM states' },
2280 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2281 AND status IN ( SELECT status_codes FROM states )
2287 Another pair of operators is C<-between> and C<-not_between>,
2288 used with an arrayref of two values:
2292 completion_date => {
2293 -not_between => ['2002-10-01', '2003-02-06']
2299 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2301 Just like with C<-in> all plausible combinations of literal SQL
2305 start0 => { -between => [ 1, 2 ] },
2306 start1 => { -between => \["? AND ?", 1, 2] },
2307 start2 => { -between => \"lower(x) AND upper(y)" },
2308 start3 => { -between => [
2310 \["upper(?)", 'stuff' ],
2317 ( start0 BETWEEN ? AND ? )
2318 AND ( start1 BETWEEN ? AND ? )
2319 AND ( start2 BETWEEN lower(x) AND upper(y) )
2320 AND ( start3 BETWEEN lower(x) AND upper(?) )
2322 @bind = (1, 2, 1, 2, 'stuff');
2325 These are the two builtin "special operators"; but the
2326 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2328 =head2 Unary operators: bool
2330 If you wish to test against boolean columns or functions within your
2331 database you can use the C<-bool> and C<-not_bool> operators. For
2332 example to test the column C<is_user> being true and the column
2333 C<is_enabled> being false you would use:-
2337 -not_bool => 'is_enabled',
2342 WHERE is_user AND NOT is_enabled
2344 If a more complex combination is required, testing more conditions,
2345 then you should use the and/or operators:-
2352 -not_bool => 'four',
2358 WHERE one AND two AND three AND NOT four
2361 =head2 Nested conditions, -and/-or prefixes
2363 So far, we've seen how multiple conditions are joined with a top-level
2364 C<AND>. We can change this by putting the different conditions we want in
2365 hashes and then putting those hashes in an array. For example:
2370 status => { -like => ['pending%', 'dispatched'] },
2374 status => 'unassigned',
2378 This data structure would create the following:
2380 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2381 OR ( user = ? AND status = ? ) )";
2382 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2385 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2386 to change the logic inside :
2392 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2393 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2400 WHERE ( user = ? AND (
2401 ( workhrs > ? AND geo = ? )
2402 OR ( workhrs < ? OR geo = ? )
2405 =head3 Algebraic inconsistency, for historical reasons
2407 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2408 operator goes C<outside> of the nested structure; whereas when connecting
2409 several constraints on one column, the C<-and> operator goes
2410 C<inside> the arrayref. Here is an example combining both features :
2413 -and => [a => 1, b => 2],
2414 -or => [c => 3, d => 4],
2415 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2420 WHERE ( ( ( a = ? AND b = ? )
2421 OR ( c = ? OR d = ? )
2422 OR ( e LIKE ? AND e LIKE ? ) ) )
2424 This difference in syntax is unfortunate but must be preserved for
2425 historical reasons. So be careful : the two examples below would
2426 seem algebraically equivalent, but they are not
2428 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2429 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2431 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2432 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2435 =head2 Literal SQL and value type operators
2437 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2438 side" is a column name and the "right side" is a value (normally rendered as
2439 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2440 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2441 alter this behavior. There are several ways of doing so.
2445 This is a virtual operator that signals the string to its right side is an
2446 identifier (a column name) and not a value. For example to compare two
2447 columns you would write:
2450 priority => { '<', 2 },
2451 requestor => { -ident => 'submitter' },
2456 $stmt = "WHERE priority < ? AND requestor = submitter";
2459 If you are maintaining legacy code you may see a different construct as
2460 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2465 This is a virtual operator that signals that the construct to its right side
2466 is a value to be passed to DBI. This is for example necessary when you want
2467 to write a where clause against an array (for RDBMS that support such
2468 datatypes). For example:
2471 array => { -value => [1, 2, 3] }
2476 $stmt = 'WHERE array = ?';
2477 @bind = ([1, 2, 3]);
2479 Note that if you were to simply say:
2485 the result would porbably be not what you wanted:
2487 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2492 Finally, sometimes only literal SQL will do. To include a random snippet
2493 of SQL verbatim, you specify it as a scalar reference. Consider this only
2494 as a last resort. Usually there is a better way. For example:
2497 priority => { '<', 2 },
2498 requestor => { -in => \'(SELECT name FROM hitmen)' },
2503 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2506 Note that in this example, you only get one bind parameter back, since
2507 the verbatim SQL is passed as part of the statement.
2511 Never use untrusted input as a literal SQL argument - this is a massive
2512 security risk (there is no way to check literal snippets for SQL
2513 injections and other nastyness). If you need to deal with untrusted input
2514 use literal SQL with placeholders as described next.
2516 =head3 Literal SQL with placeholders and bind values (subqueries)
2518 If the literal SQL to be inserted has placeholders and bind values,
2519 use a reference to an arrayref (yes this is a double reference --
2520 not so common, but perfectly legal Perl). For example, to find a date
2521 in Postgres you can use something like this:
2524 date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
2529 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2532 Note that you must pass the bind values in the same format as they are returned
2533 by L</where>. That means that if you set L</bindtype> to C<columns>, you must
2534 provide the bind values in the C<< [ column_meta => value ] >> format, where
2535 C<column_meta> is an opaque scalar value; most commonly the column name, but
2536 you can use any scalar value (including references and blessed references),
2537 L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
2538 to C<columns> the above example will look like:
2541 date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
2544 Literal SQL is especially useful for nesting parenthesized clauses in the
2545 main SQL query. Here is a first example :
2547 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2551 bar => \["IN ($sub_stmt)" => @sub_bind],
2556 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2557 WHERE c2 < ? AND c3 LIKE ?))";
2558 @bind = (1234, 100, "foo%");
2560 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2561 are expressed in the same way. Of course the C<$sub_stmt> and
2562 its associated bind values can be generated through a former call
2565 my ($sub_stmt, @sub_bind)
2566 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2567 c3 => {-like => "foo%"}});
2570 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2573 In the examples above, the subquery was used as an operator on a column;
2574 but the same principle also applies for a clause within the main C<%where>
2575 hash, like an EXISTS subquery :
2577 my ($sub_stmt, @sub_bind)
2578 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2579 my %where = ( -and => [
2581 \["EXISTS ($sub_stmt)" => @sub_bind],
2586 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2587 WHERE c1 = ? AND c2 > t0.c0))";
2591 Observe that the condition on C<c2> in the subquery refers to
2592 column C<t0.c0> of the main query : this is I<not> a bind
2593 value, so we have to express it through a scalar ref.
2594 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2595 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2596 what we wanted here.
2598 Finally, here is an example where a subquery is used
2599 for expressing unary negation:
2601 my ($sub_stmt, @sub_bind)
2602 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2603 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2605 lname => {like => '%son%'},
2606 \["NOT ($sub_stmt)" => @sub_bind],
2611 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2612 @bind = ('%son%', 10, 20)
2614 =head3 Deprecated usage of Literal SQL
2616 Below are some examples of archaic use of literal SQL. It is shown only as
2617 reference for those who deal with legacy code. Each example has a much
2618 better, cleaner and safer alternative that users should opt for in new code.
2624 my %where = ( requestor => \'IS NOT NULL' )
2626 $stmt = "WHERE requestor IS NOT NULL"
2628 This used to be the way of generating NULL comparisons, before the handling
2629 of C<undef> got formalized. For new code please use the superior syntax as
2630 described in L</Tests for NULL values>.
2634 my %where = ( requestor => \'= submitter' )
2636 $stmt = "WHERE requestor = submitter"
2638 This used to be the only way to compare columns. Use the superior L</-ident>
2639 method for all new code. For example an identifier declared in such a way
2640 will be properly quoted if L</quote_char> is properly set, while the legacy
2641 form will remain as supplied.
2645 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2647 $stmt = "WHERE completed > ? AND is_ready"
2648 @bind = ('2012-12-21')
2650 Using an empty string literal used to be the only way to express a boolean.
2651 For all new code please use the much more readable
2652 L<-bool|/Unary operators: bool> operator.
2658 These pages could go on for a while, since the nesting of the data
2659 structures this module can handle are pretty much unlimited (the
2660 module implements the C<WHERE> expansion as a recursive function
2661 internally). Your best bet is to "play around" with the module a
2662 little to see how the data structures behave, and choose the best
2663 format for your data based on that.
2665 And of course, all the values above will probably be replaced with
2666 variables gotten from forms or the command line. After all, if you
2667 knew everything ahead of time, you wouldn't have to worry about
2668 dynamically-generating SQL and could just hardwire it into your
2671 =head1 ORDER BY CLAUSES
2673 Some functions take an order by clause. This can either be a scalar (just a
2674 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2675 or an array of either of the two previous forms. Examples:
2677 Given | Will Generate
2678 ----------------------------------------------------------
2680 \'colA DESC' | ORDER BY colA DESC
2682 'colA' | ORDER BY colA
2684 [qw/colA colB/] | ORDER BY colA, colB
2686 {-asc => 'colA'} | ORDER BY colA ASC
2688 {-desc => 'colB'} | ORDER BY colB DESC
2690 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2692 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2695 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2696 { -desc => [qw/colB/], | colC ASC, colD ASC
2697 { -asc => [qw/colC colD/],|
2699 ===========================================================
2703 =head1 SPECIAL OPERATORS
2705 my $sqlmaker = SQL::Abstract->new(special_ops => [
2709 my ($self, $field, $op, $arg) = @_;
2715 handler => 'method_name',
2719 A "special operator" is a SQL syntactic clause that can be
2720 applied to a field, instead of a usual binary operator.
2723 WHERE field IN (?, ?, ?)
2724 WHERE field BETWEEN ? AND ?
2725 WHERE MATCH(field) AGAINST (?, ?)
2727 Special operators IN and BETWEEN are fairly standard and therefore
2728 are builtin within C<SQL::Abstract> (as the overridable methods
2729 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2730 like the MATCH .. AGAINST example above which is specific to MySQL,
2731 you can write your own operator handlers - supply a C<special_ops>
2732 argument to the C<new> method. That argument takes an arrayref of
2733 operator definitions; each operator definition is a hashref with two
2740 the regular expression to match the operator
2744 Either a coderef or a plain scalar method name. In both cases
2745 the expected return is C<< ($sql, @bind) >>.
2747 When supplied with a method name, it is simply called on the
2748 L<SQL::Abstract/> object as:
2750 $self->$method_name ($field, $op, $arg)
2754 $op is the part that matched the handler regex
2755 $field is the LHS of the operator
2758 When supplied with a coderef, it is called as:
2760 $coderef->($self, $field, $op, $arg)
2765 For example, here is an implementation
2766 of the MATCH .. AGAINST syntax for MySQL
2768 my $sqlmaker = SQL::Abstract->new(special_ops => [
2770 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2771 {regex => qr/^match$/i,
2773 my ($self, $field, $op, $arg) = @_;
2774 $arg = [$arg] if not ref $arg;
2775 my $label = $self->_quote($field);
2776 my ($placeholder) = $self->_convert('?');
2777 my $placeholders = join ", ", (($placeholder) x @$arg);
2778 my $sql = $self->_sqlcase('match') . " ($label) "
2779 . $self->_sqlcase('against') . " ($placeholders) ";
2780 my @bind = $self->_bindtype($field, @$arg);
2781 return ($sql, @bind);
2788 =head1 UNARY OPERATORS
2790 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2794 my ($self, $op, $arg) = @_;
2800 handler => 'method_name',
2804 A "unary operator" is a SQL syntactic clause that can be
2805 applied to a field - the operator goes before the field
2807 You can write your own operator handlers - supply a C<unary_ops>
2808 argument to the C<new> method. That argument takes an arrayref of
2809 operator definitions; each operator definition is a hashref with two
2816 the regular expression to match the operator
2820 Either a coderef or a plain scalar method name. In both cases
2821 the expected return is C<< $sql >>.
2823 When supplied with a method name, it is simply called on the
2824 L<SQL::Abstract/> object as:
2826 $self->$method_name ($op, $arg)
2830 $op is the part that matched the handler regex
2831 $arg is the RHS or argument of the operator
2833 When supplied with a coderef, it is called as:
2835 $coderef->($self, $op, $arg)
2843 Thanks to some benchmarking by Mark Stosberg, it turns out that
2844 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
2845 I must admit this wasn't an intentional design issue, but it's a
2846 byproduct of the fact that you get to control your C<DBI> handles
2849 To maximize performance, use a code snippet like the following:
2851 # prepare a statement handle using the first row
2852 # and then reuse it for the rest of the rows
2854 for my $href (@array_of_hashrefs) {
2855 $stmt ||= $sql->insert('table', $href);
2856 $sth ||= $dbh->prepare($stmt);
2857 $sth->execute($sql->values($href));
2860 The reason this works is because the keys in your C<$href> are sorted
2861 internally by B<SQL::Abstract>. Thus, as long as your data retains
2862 the same structure, you only have to generate the SQL the first time
2863 around. On subsequent queries, simply use the C<values> function provided
2864 by this module to return your values in the correct order.
2866 However this depends on the values having the same type - if, for
2867 example, the values of a where clause may either have values
2868 (resulting in sql of the form C<column = ?> with a single bind
2869 value), or alternatively the values might be C<undef> (resulting in
2870 sql of the form C<column IS NULL> with no bind value) then the
2871 caching technique suggested will not work.
2875 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
2876 really like this part (I do, at least). Building up a complex query
2877 can be as simple as the following:
2881 use CGI::FormBuilder;
2884 my $form = CGI::FormBuilder->new(...);
2885 my $sql = SQL::Abstract->new;
2887 if ($form->submitted) {
2888 my $field = $form->field;
2889 my $id = delete $field->{id};
2890 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
2893 Of course, you would still have to connect using C<DBI> to run the
2894 query, but the point is that if you make your form look like your
2895 table, the actual query script can be extremely simplistic.
2897 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
2898 a fast interface to returning and formatting data. I frequently
2899 use these three modules together to write complex database query
2900 apps in under 50 lines.
2906 =item * gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
2908 =item * git: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
2914 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
2915 Great care has been taken to preserve the I<published> behavior
2916 documented in previous versions in the 1.* family; however,
2917 some features that were previously undocumented, or behaved
2918 differently from the documentation, had to be changed in order
2919 to clarify the semantics. Hence, client code that was relying
2920 on some dark areas of C<SQL::Abstract> v1.*
2921 B<might behave differently> in v1.50.
2923 The main changes are :
2929 support for literal SQL through the C<< \ [$sql, bind] >> syntax.
2933 support for the { operator => \"..." } construct (to embed literal SQL)
2937 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
2941 optional support for L<array datatypes|/"Inserting and Updating Arrays">
2945 defensive programming : check arguments
2949 fixed bug with global logic, which was previously implemented
2950 through global variables yielding side-effects. Prior versions would
2951 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
2952 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
2953 Now this is interpreted
2954 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
2959 fixed semantics of _bindtype on array args
2963 dropped the C<_anoncopy> of the %where tree. No longer necessary,
2964 we just avoid shifting arrays within that tree.
2968 dropped the C<_modlogic> function
2972 =head1 ACKNOWLEDGEMENTS
2974 There are a number of individuals that have really helped out with
2975 this module. Unfortunately, most of them submitted bugs via CPAN
2976 so I have no idea who they are! But the people I do know are:
2978 Ash Berlin (order_by hash term support)
2979 Matt Trout (DBIx::Class support)
2980 Mark Stosberg (benchmarking)
2981 Chas Owens (initial "IN" operator support)
2982 Philip Collins (per-field SQL functions)
2983 Eric Kolve (hashref "AND" support)
2984 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
2985 Dan Kubb (support for "quote_char" and "name_sep")
2986 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
2987 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
2988 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
2989 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
2990 Oliver Charles (support for "RETURNING" after "INSERT")
2996 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3000 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3002 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3004 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3005 While not an official support venue, C<DBIx::Class> makes heavy use of
3006 C<SQL::Abstract>, and as such list members there are very familiar with
3007 how to create queries.
3011 This module is free software; you may copy this under the same
3012 terms as perl itself (either the GNU General Public License or
3013 the Artistic License)