1 package SQL::Abstract; # see doc at end of file
10 our @EXPORT_OK = qw(is_plain_value is_literal_value);
20 *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
26 #======================================================================
28 #======================================================================
30 our $VERSION = '1.79';
32 # This would confuse some packagers
33 $VERSION = eval $VERSION if $VERSION =~ /_/; # numify for warning-free dev releases
37 # special operators (-in, -between). May be extended/overridden by user.
38 # See section WHERE: BUILTIN SPECIAL OPERATORS below for implementation
39 my @BUILTIN_SPECIAL_OPS = (
40 {regex => qr/^ (?: not \s )? between $/ix, handler => '_where_field_BETWEEN'},
41 {regex => qr/^ (?: not \s )? in $/ix, handler => '_where_field_IN'},
42 {regex => qr/^ ident $/ix, handler => '_where_op_IDENT'},
43 {regex => qr/^ value $/ix, handler => '_where_op_VALUE'},
44 {regex => qr/^ is (?: \s+ not )? $/ix, handler => '_where_field_IS'},
47 # unaryish operators - key maps to handler
48 my @BUILTIN_UNARY_OPS = (
49 # the digits are backcompat stuff
50 { regex => qr/^ and (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
51 { regex => qr/^ or (?: [_\s]? \d+ )? $/xi, handler => '_where_op_ANDOR' },
52 { regex => qr/^ nest (?: [_\s]? \d+ )? $/xi, handler => '_where_op_NEST' },
53 { regex => qr/^ (?: not \s )? bool $/xi, handler => '_where_op_BOOL' },
54 { regex => qr/^ ident $/xi, handler => '_where_op_IDENT' },
55 { regex => qr/^ value $/xi, handler => '_where_op_VALUE' },
58 #======================================================================
59 # DEBUGGING AND ERROR REPORTING
60 #======================================================================
63 return unless $_[0]->{debug}; shift; # a little faster
64 my $func = (caller(1))[3];
65 warn "[$func] ", @_, "\n";
69 my($func) = (caller(1))[3];
70 Carp::carp "[$func] Warning: ", @_;
74 my($func) = (caller(1))[3];
75 Carp::croak "[$func] Fatal: ", @_;
78 sub is_literal_value ($) {
79 ref $_[0] eq 'SCALAR' ? [ ${$_[0]} ]
80 : ( ref $_[0] eq 'REF' and ref ${$_[0]} eq 'ARRAY' ) ? [ @${ $_[0] } ]
84 # FIXME XSify - this can be done so much more efficiently
85 sub is_plain_value ($) {
87 ! length ref $_[0] ? \($_[0])
89 ref $_[0] eq 'HASH' and keys %{$_[0]} == 1
91 exists $_[0]->{-value}
92 ) ? \($_[0]->{-value})
94 # reuse @_ for even moar speedz
95 defined ( $_[1] = Scalar::Util::blessed $_[0] )
97 # deliberately not using Devel::OverloadInfo - the checks we are
98 # intersted in are much more limited than the fullblown thing, and
99 # this is a very hot piece of code
101 # simply using ->can('(""') can leave behind stub methods that
102 # break actually using the overload later (see L<perldiag/Stub
103 # found while resolving method "%s" overloading "%s" in package
104 # "%s"> and the source of overload::mycan())
106 # either has stringification which DBI SHOULD prefer out of the box
107 grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
109 # has nummification or boolification, AND fallback is *not* disabled
111 SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
114 grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
116 grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
120 # no fallback specified at all
121 ! ( ($_[3]) = grep { *{"${_}::()"}{CODE} } @{$_[2]} )
123 # fallback explicitly undef
124 ! defined ${"$_[3]::()"}
137 #======================================================================
139 #======================================================================
143 my $class = ref($self) || $self;
144 my %opt = (ref $_[0] eq 'HASH') ? %{$_[0]} : @_;
146 # choose our case by keeping an option around
147 delete $opt{case} if $opt{case} && $opt{case} ne 'lower';
149 # default logic for interpreting arrayrefs
150 $opt{logic} = $opt{logic} ? uc $opt{logic} : 'OR';
152 # how to return bind vars
153 $opt{bindtype} ||= 'normal';
155 # default comparison is "=", but can be overridden
158 # try to recognize which are the 'equality' and 'inequality' ops
159 # (temporary quickfix (in 2007), should go through a more seasoned API)
160 $opt{equality_op} = qr/^( \Q$opt{cmp}\E | \= )$/ix;
161 $opt{inequality_op} = qr/^( != | <> )$/ix;
163 $opt{like_op} = qr/^ (is\s+)? r?like $/xi;
164 $opt{not_like_op} = qr/^ (is\s+)? not \s+ r?like $/xi;
167 $opt{sqltrue} ||= '1=1';
168 $opt{sqlfalse} ||= '0=1';
171 $opt{special_ops} ||= [];
172 # regexes are applied in order, thus push after user-defines
173 push @{$opt{special_ops}}, @BUILTIN_SPECIAL_OPS;
176 $opt{unary_ops} ||= [];
177 push @{$opt{unary_ops}}, @BUILTIN_UNARY_OPS;
179 # rudimentary sanity-check for user supplied bits treated as functions/operators
180 # If a purported function matches this regular expression, an exception is thrown.
181 # Literal SQL is *NOT* subject to this check, only functions (and column names
182 # when quoting is not in effect)
185 # need to guard against ()'s in column names too, but this will break tons of
186 # hacks... ideas anyone?
187 $opt{injection_guard} ||= qr/
193 return bless \%opt, $class;
197 sub _assert_pass_injection_guard {
198 if ($_[1] =~ $_[0]->{injection_guard}) {
199 my $class = ref $_[0];
200 puke "Possible SQL injection attempt '$_[1]'. If this is indeed a part of the "
201 . "desired SQL use literal SQL ( \'...' or \[ '...' ] ) or supply your own "
202 . "{injection_guard} attribute to ${class}->new()"
207 #======================================================================
209 #======================================================================
213 my $table = $self->_table(shift);
214 my $data = shift || return;
217 my $method = $self->_METHOD_FOR_refkind("_insert", $data);
218 my ($sql, @bind) = $self->$method($data);
219 $sql = join " ", $self->_sqlcase('insert into'), $table, $sql;
221 if ($options->{returning}) {
222 my ($s, @b) = $self->_insert_returning ($options);
227 return wantarray ? ($sql, @bind) : $sql;
230 sub _insert_returning {
231 my ($self, $options) = @_;
233 my $f = $options->{returning};
235 my $fieldlist = $self->_SWITCH_refkind($f, {
236 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$f;},
237 SCALAR => sub {$self->_quote($f)},
238 SCALARREF => sub {$$f},
240 return $self->_sqlcase(' returning ') . $fieldlist;
243 sub _insert_HASHREF { # explicit list of fields and then values
244 my ($self, $data) = @_;
246 my @fields = sort keys %$data;
248 my ($sql, @bind) = $self->_insert_values($data);
251 $_ = $self->_quote($_) foreach @fields;
252 $sql = "( ".join(", ", @fields).") ".$sql;
254 return ($sql, @bind);
257 sub _insert_ARRAYREF { # just generate values(?,?) part (no list of fields)
258 my ($self, $data) = @_;
260 # no names (arrayref) so can't generate bindtype
261 $self->{bindtype} ne 'columns'
262 or belch "can't do 'columns' bindtype when called with arrayref";
264 # fold the list of values into a hash of column name - value pairs
265 # (where the column names are artificially generated, and their
266 # lexicographical ordering keep the ordering of the original list)
267 my $i = "a"; # incremented values will be in lexicographical order
268 my $data_in_hash = { map { ($i++ => $_) } @$data };
270 return $self->_insert_values($data_in_hash);
273 sub _insert_ARRAYREFREF { # literal SQL with bind
274 my ($self, $data) = @_;
276 my ($sql, @bind) = @${$data};
277 $self->_assert_bindval_matches_bindtype(@bind);
279 return ($sql, @bind);
283 sub _insert_SCALARREF { # literal SQL without bind
284 my ($self, $data) = @_;
290 my ($self, $data) = @_;
292 my (@values, @all_bind);
293 foreach my $column (sort keys %$data) {
294 my $v = $data->{$column};
296 $self->_SWITCH_refkind($v, {
299 if ($self->{array_datatypes}) { # if array datatype are activated
301 push @all_bind, $self->_bindtype($column, $v);
303 else { # else literal SQL with bind
304 my ($sql, @bind) = @$v;
305 $self->_assert_bindval_matches_bindtype(@bind);
307 push @all_bind, @bind;
311 ARRAYREFREF => sub { # literal SQL with bind
312 my ($sql, @bind) = @${$v};
313 $self->_assert_bindval_matches_bindtype(@bind);
315 push @all_bind, @bind;
318 # THINK : anything useful to do with a HASHREF ?
319 HASHREF => sub { # (nothing, but old SQLA passed it through)
320 #TODO in SQLA >= 2.0 it will die instead
321 belch "HASH ref as bind value in insert is not supported";
323 push @all_bind, $self->_bindtype($column, $v);
326 SCALARREF => sub { # literal SQL without bind
330 SCALAR_or_UNDEF => sub {
332 push @all_bind, $self->_bindtype($column, $v);
339 my $sql = $self->_sqlcase('values')." ( ".join(", ", @values)." )";
340 return ($sql, @all_bind);
345 #======================================================================
347 #======================================================================
352 my $table = $self->_table(shift);
353 my $data = shift || return;
356 # first build the 'SET' part of the sql statement
357 my (@set, @all_bind);
358 puke "Unsupported data type specified to \$sql->update"
359 unless ref $data eq 'HASH';
361 for my $k (sort keys %$data) {
364 my $label = $self->_quote($k);
366 $self->_SWITCH_refkind($v, {
368 if ($self->{array_datatypes}) { # array datatype
369 push @set, "$label = ?";
370 push @all_bind, $self->_bindtype($k, $v);
372 else { # literal SQL with bind
373 my ($sql, @bind) = @$v;
374 $self->_assert_bindval_matches_bindtype(@bind);
375 push @set, "$label = $sql";
376 push @all_bind, @bind;
379 ARRAYREFREF => sub { # literal SQL with bind
380 my ($sql, @bind) = @${$v};
381 $self->_assert_bindval_matches_bindtype(@bind);
382 push @set, "$label = $sql";
383 push @all_bind, @bind;
385 SCALARREF => sub { # literal SQL without bind
386 push @set, "$label = $$v";
389 my ($op, $arg, @rest) = %$v;
391 puke 'Operator calls in update must be in the form { -op => $arg }'
392 if (@rest or not $op =~ /^\-(.+)/);
394 local $self->{_nested_func_lhs} = $k;
395 my ($sql, @bind) = $self->_where_unary_op ($1, $arg);
397 push @set, "$label = $sql";
398 push @all_bind, @bind;
400 SCALAR_or_UNDEF => sub {
401 push @set, "$label = ?";
402 push @all_bind, $self->_bindtype($k, $v);
408 my $sql = $self->_sqlcase('update') . " $table " . $self->_sqlcase('set ')
412 my($where_sql, @where_bind) = $self->where($where);
414 push @all_bind, @where_bind;
417 return wantarray ? ($sql, @all_bind) : $sql;
423 #======================================================================
425 #======================================================================
430 my $table = $self->_table(shift);
431 my $fields = shift || '*';
435 my($where_sql, @bind) = $self->where($where, $order);
437 my $f = (ref $fields eq 'ARRAY') ? join ', ', map { $self->_quote($_) } @$fields
439 my $sql = join(' ', $self->_sqlcase('select'), $f,
440 $self->_sqlcase('from'), $table)
443 return wantarray ? ($sql, @bind) : $sql;
446 #======================================================================
448 #======================================================================
453 my $table = $self->_table(shift);
457 my($where_sql, @bind) = $self->where($where);
458 my $sql = $self->_sqlcase('delete from') . " $table" . $where_sql;
460 return wantarray ? ($sql, @bind) : $sql;
464 #======================================================================
466 #======================================================================
470 # Finally, a separate routine just to handle WHERE clauses
472 my ($self, $where, $order) = @_;
475 my ($sql, @bind) = $self->_recurse_where($where);
476 $sql = $sql ? $self->_sqlcase(' where ') . "( $sql )" : '';
480 $sql .= $self->_order_by($order);
483 return wantarray ? ($sql, @bind) : $sql;
488 my ($self, $where, $logic) = @_;
490 # dispatch on appropriate method according to refkind of $where
491 my $method = $self->_METHOD_FOR_refkind("_where", $where);
493 my ($sql, @bind) = $self->$method($where, $logic);
495 # DBIx::Class used to call _recurse_where in scalar context
496 # something else might too...
498 return ($sql, @bind);
501 belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
508 #======================================================================
509 # WHERE: top-level ARRAYREF
510 #======================================================================
513 sub _where_ARRAYREF {
514 my ($self, $where, $logic) = @_;
516 $logic = uc($logic || $self->{logic});
517 $logic eq 'AND' or $logic eq 'OR' or puke "unknown logic: $logic";
519 my @clauses = @$where;
521 my (@sql_clauses, @all_bind);
522 # need to use while() so can shift() for pairs
524 my $el = shift @clauses;
526 $el = undef if (defined $el and ! length $el);
528 # switch according to kind of $el and get corresponding ($sql, @bind)
529 my ($sql, @bind) = $self->_SWITCH_refkind($el, {
531 # skip empty elements, otherwise get invalid trailing AND stuff
532 ARRAYREF => sub {$self->_recurse_where($el) if @$el},
536 $self->_assert_bindval_matches_bindtype(@b);
540 HASHREF => sub {$self->_recurse_where($el, 'and') if %$el},
542 SCALARREF => sub { ($$el); },
545 # top-level arrayref with scalars, recurse in pairs
546 $self->_recurse_where({$el => shift(@clauses)})
549 UNDEF => sub {puke "Supplying an empty left hand side argument is not supported in array-pairs" },
553 push @sql_clauses, $sql;
554 push @all_bind, @bind;
558 return $self->_join_sql_clauses($logic, \@sql_clauses, \@all_bind);
561 #======================================================================
562 # WHERE: top-level ARRAYREFREF
563 #======================================================================
565 sub _where_ARRAYREFREF {
566 my ($self, $where) = @_;
567 my ($sql, @bind) = @$$where;
568 $self->_assert_bindval_matches_bindtype(@bind);
569 return ($sql, @bind);
572 #======================================================================
573 # WHERE: top-level HASHREF
574 #======================================================================
577 my ($self, $where) = @_;
578 my (@sql_clauses, @all_bind);
580 for my $k (sort keys %$where) {
581 my $v = $where->{$k};
583 # ($k => $v) is either a special unary op or a regular hashpair
584 my ($sql, @bind) = do {
586 # put the operator in canonical form
588 $op = substr $op, 1; # remove initial dash
589 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
590 $op =~ s/\s+/ /g; # compress whitespace
592 # so that -not_foo works correctly
593 $op =~ s/^not_/NOT /i;
595 $self->_debug("Unary OP(-$op) within hashref, recursing...");
596 my ($s, @b) = $self->_where_unary_op ($op, $v);
598 # top level vs nested
599 # we assume that handled unary ops will take care of their ()s
601 List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}
603 ( defined $self->{_nested_func_lhs} and $self->{_nested_func_lhs} eq $k )
609 if (is_literal_value ($v) ) {
610 belch 'Hash-pairs consisting of an empty string with a literal are deprecated, and will be removed in 2.0: use -and => [ $literal ] instead';
613 puke "Supplying an empty left hand side argument is not supported in hash-pairs";
617 my $method = $self->_METHOD_FOR_refkind("_where_hashpair", $v);
618 $self->$method($k, $v);
622 push @sql_clauses, $sql;
623 push @all_bind, @bind;
626 return $self->_join_sql_clauses('and', \@sql_clauses, \@all_bind);
629 sub _where_unary_op {
630 my ($self, $op, $rhs) = @_;
632 if (my $op_entry = List::Util::first {$op =~ $_->{regex}} @{$self->{unary_ops}}) {
633 my $handler = $op_entry->{handler};
635 if (not ref $handler) {
636 if ($op =~ s/ [_\s]? \d+ $//x ) {
637 belch 'Use of [and|or|nest]_N modifiers is deprecated and will be removed in SQLA v2.0. '
638 . "You probably wanted ...-and => [ -$op => COND1, -$op => COND2 ... ]";
640 return $self->$handler ($op, $rhs);
642 elsif (ref $handler eq 'CODE') {
643 return $handler->($self, $op, $rhs);
646 puke "Illegal handler for operator $op - expecting a method name or a coderef";
650 $self->_debug("Generic unary OP: $op - recursing as function");
652 $self->_assert_pass_injection_guard($op);
654 my ($sql, @bind) = $self->_SWITCH_refkind ($rhs, {
656 puke "Illegal use of top-level '$op'"
657 unless defined $self->{_nested_func_lhs};
660 $self->_convert('?'),
661 $self->_bindtype($self->{_nested_func_lhs}, $rhs)
665 $self->_recurse_where ($rhs)
669 $sql = sprintf ('%s %s',
670 $self->_sqlcase($op),
674 return ($sql, @bind);
677 sub _where_op_ANDOR {
678 my ($self, $op, $v) = @_;
680 $self->_SWITCH_refkind($v, {
682 return $self->_where_ARRAYREF($v, $op);
686 return ( $op =~ /^or/i )
687 ? $self->_where_ARRAYREF( [ map { $_ => $v->{$_} } ( sort keys %$v ) ], $op )
688 : $self->_where_HASHREF($v);
692 puke "-$op => \\\$scalar makes little sense, use " .
694 ? '[ \$scalar, \%rest_of_conditions ] instead'
695 : '-and => [ \$scalar, \%rest_of_conditions ] instead'
700 puke "-$op => \\[...] makes little sense, use " .
702 ? '[ \[...], \%rest_of_conditions ] instead'
703 : '-and => [ \[...], \%rest_of_conditions ] instead'
707 SCALAR => sub { # permissively interpreted as SQL
708 puke "-$op => \$value makes little sense, use -bool => \$value instead";
712 puke "-$op => undef not supported";
718 my ($self, $op, $v) = @_;
720 $self->_SWITCH_refkind($v, {
722 SCALAR => sub { # permissively interpreted as SQL
723 belch "literal SQL should be -nest => \\'scalar' "
724 . "instead of -nest => 'scalar' ";
729 puke "-$op => undef not supported";
733 $self->_recurse_where ($v);
741 my ($self, $op, $v) = @_;
743 my ($s, @b) = $self->_SWITCH_refkind($v, {
744 SCALAR => sub { # interpreted as SQL column
745 $self->_convert($self->_quote($v));
749 puke "-$op => undef not supported";
753 $self->_recurse_where ($v);
757 $s = "(NOT $s)" if $op =~ /^not/i;
762 sub _where_op_IDENT {
764 my ($op, $rhs) = splice @_, -2;
765 if (! defined $rhs or length ref $rhs) {
766 puke "-$op requires a single plain scalar argument (a quotable identifier)";
769 # in case we are called as a top level special op (no '=')
772 $_ = $self->_convert($self->_quote($_)) for ($lhs, $rhs);
780 sub _where_op_VALUE {
782 my ($op, $rhs) = splice @_, -2;
784 # in case we are called as a top level special op (no '=')
788 if (! defined $rhs) {
790 ? $self->_convert($self->_quote($lhs)) . ' IS NULL'
797 ( defined $lhs ? $lhs : $self->{_nested_func_lhs} ),
804 $self->_convert($self->_quote($lhs)) . ' = ' . $self->_convert('?'),
808 $self->_convert('?'),
814 sub _where_hashpair_ARRAYREF {
815 my ($self, $k, $v) = @_;
818 my @v = @$v; # need copy because of shift below
819 $self->_debug("ARRAY($k) means distribute over elements");
821 # put apart first element if it is an operator (-and, -or)
823 (defined $v[0] && $v[0] =~ /^ - (?: AND|OR ) $/ix)
827 my @distributed = map { {$k => $_} } @v;
830 $self->_debug("OP($op) reinjected into the distributed array");
831 unshift @distributed, $op;
834 my $logic = $op ? substr($op, 1) : '';
836 return $self->_recurse_where(\@distributed, $logic);
839 $self->_debug("empty ARRAY($k) means 0=1");
840 return ($self->{sqlfalse});
844 sub _where_hashpair_HASHREF {
845 my ($self, $k, $v, $logic) = @_;
848 local $self->{_nested_func_lhs} = defined $self->{_nested_func_lhs}
849 ? $self->{_nested_func_lhs}
853 my ($all_sql, @all_bind);
855 for my $orig_op (sort keys %$v) {
856 my $val = $v->{$orig_op};
858 # put the operator in canonical form
861 # FIXME - we need to phase out dash-less ops
862 $op =~ s/^-//; # remove possible initial dash
863 $op =~ s/^\s+|\s+$//g;# remove leading/trailing space
864 $op =~ s/\s+/ /g; # compress whitespace
866 $self->_assert_pass_injection_guard($op);
869 $op =~ s/^is_not/IS NOT/i;
871 # so that -not_foo works correctly
872 $op =~ s/^not_/NOT /i;
874 # another retarded special case: foo => { $op => { -value => undef } }
875 if (ref $val eq 'HASH' and keys %$val == 1 and exists $val->{-value} and ! defined $val->{-value} ) {
881 # CASE: col-value logic modifiers
882 if ( $orig_op =~ /^ \- (and|or) $/xi ) {
883 ($sql, @bind) = $self->_where_hashpair_HASHREF($k, $val, $1);
885 # CASE: special operators like -in or -between
886 elsif ( my $special_op = List::Util::first {$op =~ $_->{regex}} @{$self->{special_ops}} ) {
887 my $handler = $special_op->{handler};
889 puke "No handler supplied for special operator $orig_op";
891 elsif (not ref $handler) {
892 ($sql, @bind) = $self->$handler ($k, $op, $val);
894 elsif (ref $handler eq 'CODE') {
895 ($sql, @bind) = $handler->($self, $k, $op, $val);
898 puke "Illegal handler for special operator $orig_op - expecting a method name or a coderef";
902 $self->_SWITCH_refkind($val, {
904 ARRAYREF => sub { # CASE: col => {op => \@vals}
905 ($sql, @bind) = $self->_where_field_op_ARRAYREF($k, $op, $val);
908 ARRAYREFREF => sub { # CASE: col => {op => \[$sql, @bind]} (literal SQL with bind)
909 my ($sub_sql, @sub_bind) = @$$val;
910 $self->_assert_bindval_matches_bindtype(@sub_bind);
911 $sql = join ' ', $self->_convert($self->_quote($k)),
912 $self->_sqlcase($op),
917 UNDEF => sub { # CASE: col => {op => undef} : sql "IS (NOT)? NULL"
919 $op =~ /^not$/i ? 'is not' # legacy
920 : $op =~ $self->{equality_op} ? 'is'
921 : $op =~ $self->{like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is'
922 : $op =~ $self->{inequality_op} ? 'is not'
923 : $op =~ $self->{not_like_op} ? belch("Supplying an undefined argument to '@{[ uc $op]}' is deprecated") && 'is not'
924 : puke "unexpected operator '$orig_op' with undef operand";
926 $sql = $self->_quote($k) . $self->_sqlcase(" $is null");
929 FALLBACK => sub { # CASE: col => {op/func => $stuff}
930 ($sql, @bind) = $self->_where_unary_op ($op, $val);
933 $self->_convert($self->_quote($k)),
934 $self->{_nested_func_lhs} eq $k ? $sql : "($sql)", # top level vs nested
940 ($all_sql) = (defined $all_sql and $all_sql) ? $self->_join_sql_clauses($logic, [$all_sql, $sql], []) : $sql;
941 push @all_bind, @bind;
943 return ($all_sql, @all_bind);
946 sub _where_field_IS {
947 my ($self, $k, $op, $v) = @_;
949 my ($s) = $self->_SWITCH_refkind($v, {
952 $self->_convert($self->_quote($k)),
953 map { $self->_sqlcase($_)} ($op, 'null')
956 puke "$op can only take undef as argument";
963 sub _where_field_op_ARRAYREF {
964 my ($self, $k, $op, $vals) = @_;
966 my @vals = @$vals; #always work on a copy
969 $self->_debug(sprintf '%s means multiple elements: [ %s ]',
971 join (', ', map { defined $_ ? "'$_'" : 'NULL' } @vals ),
974 # see if the first element is an -and/-or op
976 if (defined $vals[0] && $vals[0] =~ /^ - ( AND|OR ) $/ix) {
981 # a long standing API wart - an attempt to change this behavior during
982 # the 1.50 series failed *spectacularly*. Warn instead and leave the
987 (!$logic or $logic eq 'OR')
989 ( $op =~ $self->{inequality_op} or $op =~ $self->{not_like_op} )
992 belch "A multi-element arrayref as an argument to the inequality op '$o' "
993 . 'is technically equivalent to an always-true 1=1 (you probably wanted '
994 . "to say ...{ \$inequality_op => [ -and => \@values ] }... instead)"
998 # distribute $op over each remaining member of @vals, append logic if exists
999 return $self->_recurse_where([map { {$k => {$op, $_}} } @vals], $logic);
1003 # try to DWIM on equality operators
1005 $op =~ $self->{equality_op} ? $self->{sqlfalse}
1006 : $op =~ $self->{like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqlfalse}
1007 : $op =~ $self->{inequality_op} ? $self->{sqltrue}
1008 : $op =~ $self->{not_like_op} ? belch("Supplying an empty arrayref to '@{[ uc $op]}' is deprecated") && $self->{sqltrue}
1009 : puke "operator '$op' applied on an empty array (field '$k')";
1014 sub _where_hashpair_SCALARREF {
1015 my ($self, $k, $v) = @_;
1016 $self->_debug("SCALAR($k) means literal SQL: $$v");
1017 my $sql = $self->_quote($k) . " " . $$v;
1021 # literal SQL with bind
1022 sub _where_hashpair_ARRAYREFREF {
1023 my ($self, $k, $v) = @_;
1024 $self->_debug("REF($k) means literal SQL: @${$v}");
1025 my ($sql, @bind) = @$$v;
1026 $self->_assert_bindval_matches_bindtype(@bind);
1027 $sql = $self->_quote($k) . " " . $sql;
1028 return ($sql, @bind );
1031 # literal SQL without bind
1032 sub _where_hashpair_SCALAR {
1033 my ($self, $k, $v) = @_;
1034 $self->_debug("NOREF($k) means simple key=val: $k $self->{cmp} $v");
1035 my $sql = join ' ', $self->_convert($self->_quote($k)),
1036 $self->_sqlcase($self->{cmp}),
1037 $self->_convert('?');
1038 my @bind = $self->_bindtype($k, $v);
1039 return ( $sql, @bind);
1043 sub _where_hashpair_UNDEF {
1044 my ($self, $k, $v) = @_;
1045 $self->_debug("UNDEF($k) means IS NULL");
1046 my $sql = $self->_quote($k) . $self->_sqlcase(' is null');
1050 #======================================================================
1051 # WHERE: TOP-LEVEL OTHERS (SCALARREF, SCALAR, UNDEF)
1052 #======================================================================
1055 sub _where_SCALARREF {
1056 my ($self, $where) = @_;
1059 $self->_debug("SCALAR(*top) means literal SQL: $$where");
1065 my ($self, $where) = @_;
1068 $self->_debug("NOREF(*top) means literal SQL: $where");
1079 #======================================================================
1080 # WHERE: BUILTIN SPECIAL OPERATORS (-in, -between)
1081 #======================================================================
1084 sub _where_field_BETWEEN {
1085 my ($self, $k, $op, $vals) = @_;
1087 my ($label, $and, $placeholder);
1088 $label = $self->_convert($self->_quote($k));
1089 $and = ' ' . $self->_sqlcase('and') . ' ';
1090 $placeholder = $self->_convert('?');
1091 $op = $self->_sqlcase($op);
1093 my $invalid_args = "Operator '$op' requires either an arrayref with two defined values or expressions, or a single literal scalarref/arrayref-ref";
1095 my ($clause, @bind) = $self->_SWITCH_refkind($vals, {
1096 ARRAYREFREF => sub {
1097 my ($s, @b) = @$$vals;
1098 $self->_assert_bindval_matches_bindtype(@b);
1105 puke $invalid_args if @$vals != 2;
1107 my (@all_sql, @all_bind);
1108 foreach my $val (@$vals) {
1109 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1111 return ($placeholder, $self->_bindtype($k, $val) );
1116 ARRAYREFREF => sub {
1117 my ($sql, @bind) = @$$val;
1118 $self->_assert_bindval_matches_bindtype(@bind);
1119 return ($sql, @bind);
1122 my ($func, $arg, @rest) = %$val;
1123 puke ("Only simple { -func => arg } functions accepted as sub-arguments to BETWEEN")
1124 if (@rest or $func !~ /^ \- (.+)/x);
1125 $self->_where_unary_op ($1 => $arg);
1131 push @all_sql, $sql;
1132 push @all_bind, @bind;
1136 (join $and, @all_sql),
1145 my $sql = "( $label $op $clause )";
1146 return ($sql, @bind)
1150 sub _where_field_IN {
1151 my ($self, $k, $op, $vals) = @_;
1153 # backwards compatibility : if scalar, force into an arrayref
1154 $vals = [$vals] if defined $vals && ! ref $vals;
1156 my ($label) = $self->_convert($self->_quote($k));
1157 my ($placeholder) = $self->_convert('?');
1158 $op = $self->_sqlcase($op);
1160 my ($sql, @bind) = $self->_SWITCH_refkind($vals, {
1161 ARRAYREF => sub { # list of choices
1162 if (@$vals) { # nonempty list
1163 my (@all_sql, @all_bind);
1165 for my $val (@$vals) {
1166 my ($sql, @bind) = $self->_SWITCH_refkind($val, {
1168 return ($placeholder, $val);
1173 ARRAYREFREF => sub {
1174 my ($sql, @bind) = @$$val;
1175 $self->_assert_bindval_matches_bindtype(@bind);
1176 return ($sql, @bind);
1179 my ($func, $arg, @rest) = %$val;
1180 puke ("Only simple { -func => arg } functions accepted as sub-arguments to IN")
1181 if (@rest or $func !~ /^ \- (.+)/x);
1182 $self->_where_unary_op ($1 => $arg);
1186 'SQL::Abstract before v1.75 used to generate incorrect SQL when the '
1187 . "-$op operator was given an undef-containing list: !!!AUDIT YOUR CODE "
1188 . 'AND DATA!!! (the upcoming Data::Query-based version of SQL::Abstract '
1189 . 'will emit the logically correct SQL instead of raising this exception)'
1193 push @all_sql, $sql;
1194 push @all_bind, @bind;
1198 sprintf ('%s %s ( %s )',
1201 join (', ', @all_sql)
1203 $self->_bindtype($k, @all_bind),
1206 else { # empty list : some databases won't understand "IN ()", so DWIM
1207 my $sql = ($op =~ /\bnot\b/i) ? $self->{sqltrue} : $self->{sqlfalse};
1212 SCALARREF => sub { # literal SQL
1213 my $sql = $self->_open_outer_paren ($$vals);
1214 return ("$label $op ( $sql )");
1216 ARRAYREFREF => sub { # literal SQL with bind
1217 my ($sql, @bind) = @$$vals;
1218 $self->_assert_bindval_matches_bindtype(@bind);
1219 $sql = $self->_open_outer_paren ($sql);
1220 return ("$label $op ( $sql )", @bind);
1224 puke "Argument passed to the '$op' operator can not be undefined";
1228 puke "special op $op requires an arrayref (or scalarref/arrayref-ref)";
1232 return ($sql, @bind);
1235 # Some databases (SQLite) treat col IN (1, 2) different from
1236 # col IN ( (1, 2) ). Use this to strip all outer parens while
1237 # adding them back in the corresponding method
1238 sub _open_outer_paren {
1239 my ($self, $sql) = @_;
1240 $sql = $1 while $sql =~ /^ \s* \( (.*) \) \s* $/xs;
1245 #======================================================================
1247 #======================================================================
1250 my ($self, $arg) = @_;
1253 for my $c ($self->_order_by_chunks ($arg) ) {
1254 $self->_SWITCH_refkind ($c, {
1255 SCALAR => sub { push @sql, $c },
1256 ARRAYREF => sub { push @sql, shift @$c; push @bind, @$c },
1262 $self->_sqlcase(' order by'),
1268 return wantarray ? ($sql, @bind) : $sql;
1271 sub _order_by_chunks {
1272 my ($self, $arg) = @_;
1274 return $self->_SWITCH_refkind($arg, {
1277 map { $self->_order_by_chunks ($_ ) } @$arg;
1280 ARRAYREFREF => sub {
1281 my ($s, @b) = @$$arg;
1282 $self->_assert_bindval_matches_bindtype(@b);
1286 SCALAR => sub {$self->_quote($arg)},
1288 UNDEF => sub {return () },
1290 SCALARREF => sub {$$arg}, # literal SQL, no quoting
1293 # get first pair in hash
1294 my ($key, $val, @rest) = %$arg;
1296 return () unless $key;
1298 if ( @rest or not $key =~ /^-(desc|asc)/i ) {
1299 puke "hash passed to _order_by must have exactly one key (-desc or -asc)";
1305 for my $c ($self->_order_by_chunks ($val)) {
1308 $self->_SWITCH_refkind ($c, {
1313 ($sql, @bind) = @$c;
1317 $sql = $sql . ' ' . $self->_sqlcase($direction);
1319 push @ret, [ $sql, @bind];
1328 #======================================================================
1329 # DATASOURCE (FOR NOW, JUST PLAIN TABLE OR LIST OF TABLES)
1330 #======================================================================
1335 $self->_SWITCH_refkind($from, {
1336 ARRAYREF => sub {join ', ', map { $self->_quote($_) } @$from;},
1337 SCALAR => sub {$self->_quote($from)},
1338 SCALARREF => sub {$$from},
1343 #======================================================================
1345 #======================================================================
1347 # highly optimized, as it's called way too often
1349 # my ($self, $label) = @_;
1351 return '' unless defined $_[1];
1352 return ${$_[1]} if ref($_[1]) eq 'SCALAR';
1354 unless ($_[0]->{quote_char}) {
1355 $_[0]->_assert_pass_injection_guard($_[1]);
1359 my $qref = ref $_[0]->{quote_char};
1362 ($l, $r) = ( $_[0]->{quote_char}, $_[0]->{quote_char} );
1364 elsif ($qref eq 'ARRAY') {
1365 ($l, $r) = @{$_[0]->{quote_char}};
1368 puke "Unsupported quote_char format: $_[0]->{quote_char}";
1370 my $esc = $_[0]->{escape_char} || $r;
1372 # parts containing * are naturally unquoted
1373 return join( $_[0]->{name_sep}||'', map
1374 { $_ eq '*' ? $_ : do { (my $n = $_) =~ s/(\Q$esc\E|\Q$r\E)/$esc$1/g; $l . $n . $r } }
1375 ( $_[0]->{name_sep} ? split (/\Q$_[0]->{name_sep}\E/, $_[1] ) : $_[1] )
1380 # Conversion, if applicable
1382 #my ($self, $arg) = @_;
1383 if ($_[0]->{convert}) {
1384 return $_[0]->_sqlcase($_[0]->{convert}) .'(' . $_[1] . ')';
1391 #my ($self, $col, @vals) = @_;
1392 # called often - tighten code
1393 return $_[0]->{bindtype} eq 'columns'
1394 ? map {[$_[1], $_]} @_[2 .. $#_]
1399 # Dies if any element of @bind is not in [colname => value] format
1400 # if bindtype is 'columns'.
1401 sub _assert_bindval_matches_bindtype {
1402 # my ($self, @bind) = @_;
1404 if ($self->{bindtype} eq 'columns') {
1406 if (!defined $_ || ref($_) ne 'ARRAY' || @$_ != 2) {
1407 puke "bindtype 'columns' selected, you need to pass: [column_name => bind_value]"
1413 sub _join_sql_clauses {
1414 my ($self, $logic, $clauses_aref, $bind_aref) = @_;
1416 if (@$clauses_aref > 1) {
1417 my $join = " " . $self->_sqlcase($logic) . " ";
1418 my $sql = '( ' . join($join, @$clauses_aref) . ' )';
1419 return ($sql, @$bind_aref);
1421 elsif (@$clauses_aref) {
1422 return ($clauses_aref->[0], @$bind_aref); # no parentheses
1425 return (); # if no SQL, ignore @$bind_aref
1430 # Fix SQL case, if so requested
1432 # LDNOTE: if $self->{case} is true, then it contains 'lower', so we
1433 # don't touch the argument ... crooked logic, but let's not change it!
1434 return $_[0]->{case} ? $_[1] : uc($_[1]);
1438 #======================================================================
1439 # DISPATCHING FROM REFKIND
1440 #======================================================================
1443 my ($self, $data) = @_;
1445 return 'UNDEF' unless defined $data;
1447 # blessed objects are treated like scalars
1448 my $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1450 return 'SCALAR' unless $ref;
1453 while ($ref eq 'REF') {
1455 $ref = (Scalar::Util::blessed $data) ? '' : ref $data;
1459 return ($ref||'SCALAR') . ('REF' x $n_steps);
1463 my ($self, $data) = @_;
1464 my @try = ($self->_refkind($data));
1465 push @try, 'SCALAR_or_UNDEF' if $try[0] eq 'SCALAR' || $try[0] eq 'UNDEF';
1466 push @try, 'FALLBACK';
1470 sub _METHOD_FOR_refkind {
1471 my ($self, $meth_prefix, $data) = @_;
1474 for (@{$self->_try_refkind($data)}) {
1475 $method = $self->can($meth_prefix."_".$_)
1479 return $method || puke "cannot dispatch on '$meth_prefix' for ".$self->_refkind($data);
1483 sub _SWITCH_refkind {
1484 my ($self, $data, $dispatch_table) = @_;
1487 for (@{$self->_try_refkind($data)}) {
1488 $coderef = $dispatch_table->{$_}
1492 puke "no dispatch entry for ".$self->_refkind($data)
1501 #======================================================================
1502 # VALUES, GENERATE, AUTOLOAD
1503 #======================================================================
1505 # LDNOTE: original code from nwiger, didn't touch code in that section
1506 # I feel the AUTOLOAD stuff should not be the default, it should
1507 # only be activated on explicit demand by user.
1511 my $data = shift || return;
1512 puke "Argument to ", __PACKAGE__, "->values must be a \\%hash"
1513 unless ref $data eq 'HASH';
1516 foreach my $k ( sort keys %$data ) {
1517 my $v = $data->{$k};
1518 $self->_SWITCH_refkind($v, {
1520 if ($self->{array_datatypes}) { # array datatype
1521 push @all_bind, $self->_bindtype($k, $v);
1523 else { # literal SQL with bind
1524 my ($sql, @bind) = @$v;
1525 $self->_assert_bindval_matches_bindtype(@bind);
1526 push @all_bind, @bind;
1529 ARRAYREFREF => sub { # literal SQL with bind
1530 my ($sql, @bind) = @${$v};
1531 $self->_assert_bindval_matches_bindtype(@bind);
1532 push @all_bind, @bind;
1534 SCALARREF => sub { # literal SQL without bind
1536 SCALAR_or_UNDEF => sub {
1537 push @all_bind, $self->_bindtype($k, $v);
1548 my(@sql, @sqlq, @sqlv);
1552 if ($ref eq 'HASH') {
1553 for my $k (sort keys %$_) {
1556 my $label = $self->_quote($k);
1557 if ($r eq 'ARRAY') {
1558 # literal SQL with bind
1559 my ($sql, @bind) = @$v;
1560 $self->_assert_bindval_matches_bindtype(@bind);
1561 push @sqlq, "$label = $sql";
1563 } elsif ($r eq 'SCALAR') {
1564 # literal SQL without bind
1565 push @sqlq, "$label = $$v";
1567 push @sqlq, "$label = ?";
1568 push @sqlv, $self->_bindtype($k, $v);
1571 push @sql, $self->_sqlcase('set'), join ', ', @sqlq;
1572 } elsif ($ref eq 'ARRAY') {
1573 # unlike insert(), assume these are ONLY the column names, i.e. for SQL
1576 if ($r eq 'ARRAY') { # literal SQL with bind
1577 my ($sql, @bind) = @$v;
1578 $self->_assert_bindval_matches_bindtype(@bind);
1581 } elsif ($r eq 'SCALAR') { # literal SQL without bind
1582 # embedded literal SQL
1589 push @sql, '(' . join(', ', @sqlq) . ')';
1590 } elsif ($ref eq 'SCALAR') {
1594 # strings get case twiddled
1595 push @sql, $self->_sqlcase($_);
1599 my $sql = join ' ', @sql;
1601 # this is pretty tricky
1602 # if ask for an array, return ($stmt, @bind)
1603 # otherwise, s/?/shift @sqlv/ to put it inline
1605 return ($sql, @sqlv);
1607 1 while $sql =~ s/\?/my $d = shift(@sqlv);
1608 ref $d ? $d->[1] : $d/e;
1617 # This allows us to check for a local, then _form, attr
1619 my($name) = $AUTOLOAD =~ /.*::(.+)/;
1620 return $self->generate($name, @_);
1631 SQL::Abstract - Generate SQL from Perl data structures
1637 my $sql = SQL::Abstract->new;
1639 my($stmt, @bind) = $sql->select($source, \@fields, \%where, \@order);
1641 my($stmt, @bind) = $sql->insert($table, \%fieldvals || \@values);
1643 my($stmt, @bind) = $sql->update($table, \%fieldvals, \%where);
1645 my($stmt, @bind) = $sql->delete($table, \%where);
1647 # Then, use these in your DBI statements
1648 my $sth = $dbh->prepare($stmt);
1649 $sth->execute(@bind);
1651 # Just generate the WHERE clause
1652 my($stmt, @bind) = $sql->where(\%where, \@order);
1654 # Return values in the same order, for hashed queries
1655 # See PERFORMANCE section for more details
1656 my @bind = $sql->values(\%fieldvals);
1660 This module was inspired by the excellent L<DBIx::Abstract>.
1661 However, in using that module I found that what I really wanted
1662 to do was generate SQL, but still retain complete control over my
1663 statement handles and use the DBI interface. So, I set out to
1664 create an abstract SQL generation module.
1666 While based on the concepts used by L<DBIx::Abstract>, there are
1667 several important differences, especially when it comes to WHERE
1668 clauses. I have modified the concepts used to make the SQL easier
1669 to generate from Perl data structures and, IMO, more intuitive.
1670 The underlying idea is for this module to do what you mean, based
1671 on the data structures you provide it. The big advantage is that
1672 you don't have to modify your code every time your data changes,
1673 as this module figures it out.
1675 To begin with, an SQL INSERT is as easy as just specifying a hash
1676 of C<key=value> pairs:
1679 name => 'Jimbo Bobson',
1680 phone => '123-456-7890',
1681 address => '42 Sister Lane',
1682 city => 'St. Louis',
1683 state => 'Louisiana',
1686 The SQL can then be generated with this:
1688 my($stmt, @bind) = $sql->insert('people', \%data);
1690 Which would give you something like this:
1692 $stmt = "INSERT INTO people
1693 (address, city, name, phone, state)
1694 VALUES (?, ?, ?, ?, ?)";
1695 @bind = ('42 Sister Lane', 'St. Louis', 'Jimbo Bobson',
1696 '123-456-7890', 'Louisiana');
1698 These are then used directly in your DBI code:
1700 my $sth = $dbh->prepare($stmt);
1701 $sth->execute(@bind);
1703 =head2 Inserting and Updating Arrays
1705 If your database has array types (like for example Postgres),
1706 activate the special option C<< array_datatypes => 1 >>
1707 when creating the C<SQL::Abstract> object.
1708 Then you may use an arrayref to insert and update database array types:
1710 my $sql = SQL::Abstract->new(array_datatypes => 1);
1712 planets => [qw/Mercury Venus Earth Mars/]
1715 my($stmt, @bind) = $sql->insert('solar_system', \%data);
1719 $stmt = "INSERT INTO solar_system (planets) VALUES (?)"
1721 @bind = (['Mercury', 'Venus', 'Earth', 'Mars']);
1724 =head2 Inserting and Updating SQL
1726 In order to apply SQL functions to elements of your C<%data> you may
1727 specify a reference to an arrayref for the given hash value. For example,
1728 if you need to execute the Oracle C<to_date> function on a value, you can
1729 say something like this:
1733 date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
1736 The first value in the array is the actual SQL. Any other values are
1737 optional and would be included in the bind values array. This gives
1740 my($stmt, @bind) = $sql->insert('people', \%data);
1742 $stmt = "INSERT INTO people (name, date_entered)
1743 VALUES (?, to_date(?,'MM/DD/YYYY'))";
1744 @bind = ('Bill', '03/02/2003');
1746 An UPDATE is just as easy, all you change is the name of the function:
1748 my($stmt, @bind) = $sql->update('people', \%data);
1750 Notice that your C<%data> isn't touched; the module will generate
1751 the appropriately quirky SQL for you automatically. Usually you'll
1752 want to specify a WHERE clause for your UPDATE, though, which is
1753 where handling C<%where> hashes comes in handy...
1755 =head2 Complex where statements
1757 This module can generate pretty complicated WHERE statements
1758 easily. For example, simple C<key=value> pairs are taken to mean
1759 equality, and if you want to see if a field is within a set
1760 of values, you can use an arrayref. Let's say we wanted to
1761 SELECT some data based on this criteria:
1764 requestor => 'inna',
1765 worker => ['nwiger', 'rcwe', 'sfz'],
1766 status => { '!=', 'completed' }
1769 my($stmt, @bind) = $sql->select('tickets', '*', \%where);
1771 The above would give you something like this:
1773 $stmt = "SELECT * FROM tickets WHERE
1774 ( requestor = ? ) AND ( status != ? )
1775 AND ( worker = ? OR worker = ? OR worker = ? )";
1776 @bind = ('inna', 'completed', 'nwiger', 'rcwe', 'sfz');
1778 Which you could then use in DBI code like so:
1780 my $sth = $dbh->prepare($stmt);
1781 $sth->execute(@bind);
1787 The methods are simple. There's one for every major SQL operation,
1788 and a constructor you use first. The arguments are specified in a
1789 similar order for each method (table, then fields, then a where
1790 clause) to try and simplify things.
1792 =head2 new(option => 'value')
1794 The C<new()> function takes a list of options and values, and returns
1795 a new B<SQL::Abstract> object which can then be used to generate SQL
1796 through the methods below. The options accepted are:
1802 If set to 'lower', then SQL will be generated in all lowercase. By
1803 default SQL is generated in "textbook" case meaning something like:
1805 SELECT a_field FROM a_table WHERE some_field LIKE '%someval%'
1807 Any setting other than 'lower' is ignored.
1811 This determines what the default comparison operator is. By default
1812 it is C<=>, meaning that a hash like this:
1814 %where = (name => 'nwiger', email => 'nate@wiger.org');
1816 Will generate SQL like this:
1818 WHERE name = 'nwiger' AND email = 'nate@wiger.org'
1820 However, you may want loose comparisons by default, so if you set
1821 C<cmp> to C<like> you would get SQL such as:
1823 WHERE name like 'nwiger' AND email like 'nate@wiger.org'
1825 You can also override the comparison on an individual basis - see
1826 the huge section on L</"WHERE CLAUSES"> at the bottom.
1828 =item sqltrue, sqlfalse
1830 Expressions for inserting boolean values within SQL statements.
1831 By default these are C<1=1> and C<1=0>. They are used
1832 by the special operators C<-in> and C<-not_in> for generating
1833 correct SQL even when the argument is an empty array (see below).
1837 This determines the default logical operator for multiple WHERE
1838 statements in arrays or hashes. If absent, the default logic is "or"
1839 for arrays, and "and" for hashes. This means that a WHERE
1843 event_date => {'>=', '2/13/99'},
1844 event_date => {'<=', '4/24/03'},
1847 will generate SQL like this:
1849 WHERE event_date >= '2/13/99' OR event_date <= '4/24/03'
1851 This is probably not what you want given this query, though (look
1852 at the dates). To change the "OR" to an "AND", simply specify:
1854 my $sql = SQL::Abstract->new(logic => 'and');
1856 Which will change the above C<WHERE> to:
1858 WHERE event_date >= '2/13/99' AND event_date <= '4/24/03'
1860 The logic can also be changed locally by inserting
1861 a modifier in front of an arrayref :
1863 @where = (-and => [event_date => {'>=', '2/13/99'},
1864 event_date => {'<=', '4/24/03'} ]);
1866 See the L</"WHERE CLAUSES"> section for explanations.
1870 This will automatically convert comparisons using the specified SQL
1871 function for both column and value. This is mostly used with an argument
1872 of C<upper> or C<lower>, so that the SQL will have the effect of
1873 case-insensitive "searches". For example, this:
1875 $sql = SQL::Abstract->new(convert => 'upper');
1876 %where = (keywords => 'MaKe iT CAse inSeNSItive');
1878 Will turn out the following SQL:
1880 WHERE upper(keywords) like upper('MaKe iT CAse inSeNSItive')
1882 The conversion can be C<upper()>, C<lower()>, or any other SQL function
1883 that can be applied symmetrically to fields (actually B<SQL::Abstract> does
1884 not validate this option; it will just pass through what you specify verbatim).
1888 This is a kludge because many databases suck. For example, you can't
1889 just bind values using DBI's C<execute()> for Oracle C<CLOB> or C<BLOB> fields.
1890 Instead, you have to use C<bind_param()>:
1892 $sth->bind_param(1, 'reg data');
1893 $sth->bind_param(2, $lots, {ora_type => ORA_CLOB});
1895 The problem is, B<SQL::Abstract> will normally just return a C<@bind> array,
1896 which loses track of which field each slot refers to. Fear not.
1898 If you specify C<bindtype> in new, you can determine how C<@bind> is returned.
1899 Currently, you can specify either C<normal> (default) or C<columns>. If you
1900 specify C<columns>, you will get an array that looks like this:
1902 my $sql = SQL::Abstract->new(bindtype => 'columns');
1903 my($stmt, @bind) = $sql->insert(...);
1906 [ 'column1', 'value1' ],
1907 [ 'column2', 'value2' ],
1908 [ 'column3', 'value3' ],
1911 You can then iterate through this manually, using DBI's C<bind_param()>.
1913 $sth->prepare($stmt);
1916 my($col, $data) = @$_;
1917 if ($col eq 'details' || $col eq 'comments') {
1918 $sth->bind_param($i, $data, {ora_type => ORA_CLOB});
1919 } elsif ($col eq 'image') {
1920 $sth->bind_param($i, $data, {ora_type => ORA_BLOB});
1922 $sth->bind_param($i, $data);
1926 $sth->execute; # execute without @bind now
1928 Now, why would you still use B<SQL::Abstract> if you have to do this crap?
1929 Basically, the advantage is still that you don't have to care which fields
1930 are or are not included. You could wrap that above C<for> loop in a simple
1931 sub called C<bind_fields()> or something and reuse it repeatedly. You still
1932 get a layer of abstraction over manual SQL specification.
1934 Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
1935 construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
1936 will expect the bind values in this format.
1940 This is the character that a table or column name will be quoted
1941 with. By default this is an empty string, but you could set it to
1942 the character C<`>, to generate SQL like this:
1944 SELECT `a_field` FROM `a_table` WHERE `some_field` LIKE '%someval%'
1946 Alternatively, you can supply an array ref of two items, the first being the left
1947 hand quote character, and the second the right hand quote character. For
1948 example, you could supply C<['[',']']> for SQL Server 2000 compliant quotes
1949 that generates SQL like this:
1951 SELECT [a_field] FROM [a_table] WHERE [some_field] LIKE '%someval%'
1953 Quoting is useful if you have tables or columns names that are reserved
1954 words in your database's SQL dialect.
1958 This is the character that will be used to escape L</quote_char>s appearing
1959 in an identifier before it has been quoted.
1961 The parameter default in case of a single L</quote_char> character is the quote
1964 When opening-closing-style quoting is used (L</quote_char> is an arrayref)
1965 this parameter defaults to the B<closing (right)> L</quote_char>. Occurences
1966 of the B<opening (left)> L</quote_char> within the identifier are currently left
1967 untouched. The default for opening-closing-style quotes may change in future
1968 versions, thus you are B<strongly encouraged> to specify the escape character
1973 This is the character that separates a table and column name. It is
1974 necessary to specify this when the C<quote_char> option is selected,
1975 so that tables and column names can be individually quoted like this:
1977 SELECT `table`.`one_field` FROM `table` WHERE `table`.`other_field` = 1
1979 =item injection_guard
1981 A regular expression C<qr/.../> that is applied to any C<-function> and unquoted
1982 column name specified in a query structure. This is a safety mechanism to avoid
1983 injection attacks when mishandling user input e.g.:
1985 my %condition_as_column_value_pairs = get_values_from_user();
1986 $sqla->select( ... , \%condition_as_column_value_pairs );
1988 If the expression matches an exception is thrown. Note that literal SQL
1989 supplied via C<\'...'> or C<\['...']> is B<not> checked in any way.
1991 Defaults to checking for C<;> and the C<GO> keyword (TransactSQL)
1993 =item array_datatypes
1995 When this option is true, arrayrefs in INSERT or UPDATE are
1996 interpreted as array datatypes and are passed directly
1998 When this option is false, arrayrefs are interpreted
1999 as literal SQL, just like refs to arrayrefs
2000 (but this behavior is for backwards compatibility; when writing
2001 new queries, use the "reference to arrayref" syntax
2007 Takes a reference to a list of "special operators"
2008 to extend the syntax understood by L<SQL::Abstract>.
2009 See section L</"SPECIAL OPERATORS"> for details.
2013 Takes a reference to a list of "unary operators"
2014 to extend the syntax understood by L<SQL::Abstract>.
2015 See section L</"UNARY OPERATORS"> for details.
2021 =head2 insert($table, \@values || \%fieldvals, \%options)
2023 This is the simplest function. You simply give it a table name
2024 and either an arrayref of values or hashref of field/value pairs.
2025 It returns an SQL INSERT statement and a list of bind values.
2026 See the sections on L</"Inserting and Updating Arrays"> and
2027 L</"Inserting and Updating SQL"> for information on how to insert
2028 with those data types.
2030 The optional C<\%options> hash reference may contain additional
2031 options to generate the insert SQL. Currently supported options
2038 Takes either a scalar of raw SQL fields, or an array reference of
2039 field names, and adds on an SQL C<RETURNING> statement at the end.
2040 This allows you to return data generated by the insert statement
2041 (such as row IDs) without performing another C<SELECT> statement.
2042 Note, however, this is not part of the SQL standard and may not
2043 be supported by all database engines.
2047 =head2 update($table, \%fieldvals, \%where)
2049 This takes a table, hashref of field/value pairs, and an optional
2050 hashref L<WHERE clause|/WHERE CLAUSES>. It returns an SQL UPDATE function and a list
2052 See the sections on L</"Inserting and Updating Arrays"> and
2053 L</"Inserting and Updating SQL"> for information on how to insert
2054 with those data types.
2056 =head2 select($source, $fields, $where, $order)
2058 This returns a SQL SELECT statement and associated list of bind values, as
2059 specified by the arguments :
2065 Specification of the 'FROM' part of the statement.
2066 The argument can be either a plain scalar (interpreted as a table
2067 name, will be quoted), or an arrayref (interpreted as a list
2068 of table names, joined by commas, quoted), or a scalarref
2069 (literal table name, not quoted), or a ref to an arrayref
2070 (list of literal table names, joined by commas, not quoted).
2074 Specification of the list of fields to retrieve from
2076 The argument can be either an arrayref (interpreted as a list
2077 of field names, will be joined by commas and quoted), or a
2078 plain scalar (literal SQL, not quoted).
2079 Please observe that this API is not as flexible as that of
2080 the first argument C<$source>, for backwards compatibility reasons.
2084 Optional argument to specify the WHERE part of the query.
2085 The argument is most often a hashref, but can also be
2086 an arrayref or plain scalar --
2087 see section L<WHERE clause|/"WHERE CLAUSES"> for details.
2091 Optional argument to specify the ORDER BY part of the query.
2092 The argument can be a scalar, a hashref or an arrayref
2093 -- see section L<ORDER BY clause|/"ORDER BY CLAUSES">
2099 =head2 delete($table, \%where)
2101 This takes a table name and optional hashref L<WHERE clause|/WHERE CLAUSES>.
2102 It returns an SQL DELETE statement and list of bind values.
2104 =head2 where(\%where, \@order)
2106 This is used to generate just the WHERE clause. For example,
2107 if you have an arbitrary data structure and know what the
2108 rest of your SQL is going to look like, but want an easy way
2109 to produce a WHERE clause, use this. It returns an SQL WHERE
2110 clause and list of bind values.
2113 =head2 values(\%data)
2115 This just returns the values from the hash C<%data>, in the same
2116 order that would be returned from any of the other above queries.
2117 Using this allows you to markedly speed up your queries if you
2118 are affecting lots of rows. See below under the L</"PERFORMANCE"> section.
2120 =head2 generate($any, 'number', $of, \@data, $struct, \%types)
2122 Warning: This is an experimental method and subject to change.
2124 This returns arbitrarily generated SQL. It's a really basic shortcut.
2125 It will return two different things, depending on return context:
2127 my($stmt, @bind) = $sql->generate('create table', \$table, \@fields);
2128 my $stmt_and_val = $sql->generate('create table', \$table, \@fields);
2130 These would return the following:
2132 # First calling form
2133 $stmt = "CREATE TABLE test (?, ?)";
2134 @bind = (field1, field2);
2136 # Second calling form
2137 $stmt_and_val = "CREATE TABLE test (field1, field2)";
2139 Depending on what you're trying to do, it's up to you to choose the correct
2140 format. In this example, the second form is what you would want.
2144 $sql->generate('alter session', { nls_date_format => 'MM/YY' });
2148 ALTER SESSION SET nls_date_format = 'MM/YY'
2150 You get the idea. Strings get their case twiddled, but everything
2151 else remains verbatim.
2153 =head1 EXPORTABLE FUNCTIONS
2155 =head2 is_plain_value
2157 Determines if the supplied argument is a plain value as understood by this
2162 =item * The value is C<undef>
2164 =item * The value is a non-reference
2166 =item * The value is an object with stringification overloading
2168 =item * The value is of the form C<< { -value => $anything } >>
2172 On failure returns C<undef>, on sucess returns a B<scalar> reference
2173 to the original supplied argument.
2179 The stringification overloading detection is rather advanced: it takes
2180 into consideration not only the presence of a C<""> overload, but if that
2181 fails also checks for enabled
2182 L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
2183 on either C<0+> or C<bool>.
2185 Unfortunately testing in the field indicates that this
2186 detection B<< may tickle a latent bug in perl versions before 5.018 >>,
2187 but only when very large numbers of stringifying objects are involved.
2188 At the time of writing ( Sep 2014 ) there is no clear explanation of
2189 the direct cause, nor is there a manageably small test case that reliably
2190 reproduces the problem.
2192 If you encounter any of the following exceptions in B<random places within
2193 your application stack> - this module may be to blame:
2195 Operation "ne": no method found,
2196 left argument in overloaded package <something>,
2197 right argument in overloaded package <something>
2201 Stub found while resolving method "???" overloading """" in package <something>
2203 If you fall victim to the above - please attempt to reduce the problem
2204 to something that could be sent to the L<SQL::Abstract developers
2205 |DBIx::Class/GETTING HELP/SUPPORT>
2206 (either publicly or privately). As a workaround in the meantime you can
2207 set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
2208 value, which will most likely eliminate your problem (at the expense of
2209 not being able to properly detect exotic forms of stringification).
2211 This notice and environment variable will be removed in a future version,
2212 as soon as the underlying problem is found and a reliable workaround is
2217 =head2 is_literal_value
2219 Determines if the supplied argument is a literal value as understood by this
2224 =item * C<\$sql_string>
2226 =item * C<\[ $sql_string, @bind_values ]>
2230 On failure returns C<undef>, on sucess returns an B<array> reference
2231 containing the unpacked version of the supplied literal SQL and bind values.
2233 =head1 WHERE CLAUSES
2237 This module uses a variation on the idea from L<DBIx::Abstract>. It
2238 is B<NOT>, repeat I<not> 100% compatible. B<The main logic of this
2239 module is that things in arrays are OR'ed, and things in hashes
2242 The easiest way to explain is to show lots of examples. After
2243 each C<%where> hash shown, it is assumed you used:
2245 my($stmt, @bind) = $sql->where(\%where);
2247 However, note that the C<%where> hash can be used directly in any
2248 of the other functions as well, as described above.
2250 =head2 Key-value pairs
2252 So, let's get started. To begin, a simple hash:
2256 status => 'completed'
2259 Is converted to SQL C<key = val> statements:
2261 $stmt = "WHERE user = ? AND status = ?";
2262 @bind = ('nwiger', 'completed');
2264 One common thing I end up doing is having a list of values that
2265 a field can be in. To do this, simply specify a list inside of
2270 status => ['assigned', 'in-progress', 'pending'];
2273 This simple code will create the following:
2275 $stmt = "WHERE user = ? AND ( status = ? OR status = ? OR status = ? )";
2276 @bind = ('nwiger', 'assigned', 'in-progress', 'pending');
2278 A field associated to an empty arrayref will be considered a
2279 logical false and will generate 0=1.
2281 =head2 Tests for NULL values
2283 If the value part is C<undef> then this is converted to SQL <IS NULL>
2292 $stmt = "WHERE user = ? AND status IS NULL";
2295 To test if a column IS NOT NULL:
2299 status => { '!=', undef },
2302 =head2 Specific comparison operators
2304 If you want to specify a different type of operator for your comparison,
2305 you can use a hashref for a given column:
2309 status => { '!=', 'completed' }
2312 Which would generate:
2314 $stmt = "WHERE user = ? AND status != ?";
2315 @bind = ('nwiger', 'completed');
2317 To test against multiple values, just enclose the values in an arrayref:
2319 status => { '=', ['assigned', 'in-progress', 'pending'] };
2321 Which would give you:
2323 "WHERE status = ? OR status = ? OR status = ?"
2326 The hashref can also contain multiple pairs, in which case it is expanded
2327 into an C<AND> of its elements:
2331 status => { '!=', 'completed', -not_like => 'pending%' }
2334 # Or more dynamically, like from a form
2335 $where{user} = 'nwiger';
2336 $where{status}{'!='} = 'completed';
2337 $where{status}{'-not_like'} = 'pending%';
2339 # Both generate this
2340 $stmt = "WHERE user = ? AND status != ? AND status NOT LIKE ?";
2341 @bind = ('nwiger', 'completed', 'pending%');
2344 To get an OR instead, you can combine it with the arrayref idea:
2348 priority => [ { '=', 2 }, { '>', 5 } ]
2351 Which would generate:
2353 $stmt = "WHERE ( priority = ? OR priority > ? ) AND user = ?";
2354 @bind = ('2', '5', 'nwiger');
2356 If you want to include literal SQL (with or without bind values), just use a
2357 scalar reference or reference to an arrayref as the value:
2360 date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
2361 date_expires => { '<' => \"now()" }
2364 Which would generate:
2366 $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
2367 @bind = ('11/26/2008');
2370 =head2 Logic and nesting operators
2372 In the example above,
2373 there is a subtle trap if you want to say something like
2374 this (notice the C<AND>):
2376 WHERE priority != ? AND priority != ?
2378 Because, in Perl you I<can't> do this:
2380 priority => { '!=' => 2, '!=' => 1 }
2382 As the second C<!=> key will obliterate the first. The solution
2383 is to use the special C<-modifier> form inside an arrayref:
2385 priority => [ -and => {'!=', 2},
2389 Normally, these would be joined by C<OR>, but the modifier tells it
2390 to use C<AND> instead. (Hint: You can use this in conjunction with the
2391 C<logic> option to C<new()> in order to change the way your queries
2392 work by default.) B<Important:> Note that the C<-modifier> goes
2393 B<INSIDE> the arrayref, as an extra first element. This will
2394 B<NOT> do what you think it might:
2396 priority => -and => [{'!=', 2}, {'!=', 1}] # WRONG!
2398 Here is a quick list of equivalencies, since there is some overlap:
2401 status => {'!=', 'completed', 'not like', 'pending%' }
2402 status => [ -and => {'!=', 'completed'}, {'not like', 'pending%'}]
2405 status => {'=', ['assigned', 'in-progress']}
2406 status => [ -or => {'=', 'assigned'}, {'=', 'in-progress'}]
2407 status => [ {'=', 'assigned'}, {'=', 'in-progress'} ]
2411 =head2 Special operators : IN, BETWEEN, etc.
2413 You can also use the hashref format to compare a list of fields using the
2414 C<IN> comparison operator, by specifying the list as an arrayref:
2417 status => 'completed',
2418 reportid => { -in => [567, 2335, 2] }
2421 Which would generate:
2423 $stmt = "WHERE status = ? AND reportid IN (?,?,?)";
2424 @bind = ('completed', '567', '2335', '2');
2426 The reverse operator C<-not_in> generates SQL C<NOT IN> and is used in
2429 If the argument to C<-in> is an empty array, 'sqlfalse' is generated
2430 (by default : C<1=0>). Similarly, C<< -not_in => [] >> generates
2431 'sqltrue' (by default : C<1=1>).
2433 In addition to the array you can supply a chunk of literal sql or
2434 literal sql with bind:
2437 customer => { -in => \[
2438 'SELECT cust_id FROM cust WHERE balance > ?',
2441 status => { -in => \'SELECT status_codes FROM states' },
2447 customer IN ( SELECT cust_id FROM cust WHERE balance > ? )
2448 AND status IN ( SELECT status_codes FROM states )
2452 Finally, if the argument to C<-in> is not a reference, it will be
2453 treated as a single-element array.
2455 Another pair of operators is C<-between> and C<-not_between>,
2456 used with an arrayref of two values:
2460 completion_date => {
2461 -not_between => ['2002-10-01', '2003-02-06']
2467 WHERE user = ? AND completion_date NOT BETWEEN ( ? AND ? )
2469 Just like with C<-in> all plausible combinations of literal SQL
2473 start0 => { -between => [ 1, 2 ] },
2474 start1 => { -between => \["? AND ?", 1, 2] },
2475 start2 => { -between => \"lower(x) AND upper(y)" },
2476 start3 => { -between => [
2478 \["upper(?)", 'stuff' ],
2485 ( start0 BETWEEN ? AND ? )
2486 AND ( start1 BETWEEN ? AND ? )
2487 AND ( start2 BETWEEN lower(x) AND upper(y) )
2488 AND ( start3 BETWEEN lower(x) AND upper(?) )
2490 @bind = (1, 2, 1, 2, 'stuff');
2493 These are the two builtin "special operators"; but the
2494 list can be expanded : see section L</"SPECIAL OPERATORS"> below.
2496 =head2 Unary operators: bool
2498 If you wish to test against boolean columns or functions within your
2499 database you can use the C<-bool> and C<-not_bool> operators. For
2500 example to test the column C<is_user> being true and the column
2501 C<is_enabled> being false you would use:-
2505 -not_bool => 'is_enabled',
2510 WHERE is_user AND NOT is_enabled
2512 If a more complex combination is required, testing more conditions,
2513 then you should use the and/or operators:-
2518 -not_bool => { two=> { -rlike => 'bar' } },
2519 -not_bool => { three => [ { '=', 2 }, { '>', 5 } ] },
2530 (NOT ( three = ? OR three > ? ))
2533 =head2 Nested conditions, -and/-or prefixes
2535 So far, we've seen how multiple conditions are joined with a top-level
2536 C<AND>. We can change this by putting the different conditions we want in
2537 hashes and then putting those hashes in an array. For example:
2542 status => { -like => ['pending%', 'dispatched'] },
2546 status => 'unassigned',
2550 This data structure would create the following:
2552 $stmt = "WHERE ( user = ? AND ( status LIKE ? OR status LIKE ? ) )
2553 OR ( user = ? AND status = ? ) )";
2554 @bind = ('nwiger', 'pending', 'dispatched', 'robot', 'unassigned');
2557 Clauses in hashrefs or arrayrefs can be prefixed with an C<-and> or C<-or>
2558 to change the logic inside :
2564 -and => [ workhrs => {'>', 20}, geo => 'ASIA' ],
2565 -or => { workhrs => {'<', 50}, geo => 'EURO' },
2572 $stmt = "WHERE ( user = ?
2573 AND ( ( workhrs > ? AND geo = ? )
2574 OR ( workhrs < ? OR geo = ? ) ) )";
2575 @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
2577 =head3 Algebraic inconsistency, for historical reasons
2579 C<Important note>: when connecting several conditions, the C<-and->|C<-or>
2580 operator goes C<outside> of the nested structure; whereas when connecting
2581 several constraints on one column, the C<-and> operator goes
2582 C<inside> the arrayref. Here is an example combining both features :
2585 -and => [a => 1, b => 2],
2586 -or => [c => 3, d => 4],
2587 e => [-and => {-like => 'foo%'}, {-like => '%bar'} ]
2592 WHERE ( ( ( a = ? AND b = ? )
2593 OR ( c = ? OR d = ? )
2594 OR ( e LIKE ? AND e LIKE ? ) ) )
2596 This difference in syntax is unfortunate but must be preserved for
2597 historical reasons. So be careful : the two examples below would
2598 seem algebraically equivalent, but they are not
2600 {col => [-and => {-like => 'foo%'}, {-like => '%bar'}]}
2601 # yields : WHERE ( ( col LIKE ? AND col LIKE ? ) )
2603 [-and => {col => {-like => 'foo%'}, {col => {-like => '%bar'}}]]
2604 # yields : WHERE ( ( col LIKE ? OR col LIKE ? ) )
2607 =head2 Literal SQL and value type operators
2609 The basic premise of SQL::Abstract is that in WHERE specifications the "left
2610 side" is a column name and the "right side" is a value (normally rendered as
2611 a placeholder). This holds true for both hashrefs and arrayref pairs as you
2612 see in the L</WHERE CLAUSES> examples above. Sometimes it is necessary to
2613 alter this behavior. There are several ways of doing so.
2617 This is a virtual operator that signals the string to its right side is an
2618 identifier (a column name) and not a value. For example to compare two
2619 columns you would write:
2622 priority => { '<', 2 },
2623 requestor => { -ident => 'submitter' },
2628 $stmt = "WHERE priority < ? AND requestor = submitter";
2631 If you are maintaining legacy code you may see a different construct as
2632 described in L</Deprecated usage of Literal SQL>, please use C<-ident> in new
2637 This is a virtual operator that signals that the construct to its right side
2638 is a value to be passed to DBI. This is for example necessary when you want
2639 to write a where clause against an array (for RDBMS that support such
2640 datatypes). For example:
2643 array => { -value => [1, 2, 3] }
2648 $stmt = 'WHERE array = ?';
2649 @bind = ([1, 2, 3]);
2651 Note that if you were to simply say:
2657 the result would probably not be what you wanted:
2659 $stmt = 'WHERE array = ? OR array = ? OR array = ?';
2664 Finally, sometimes only literal SQL will do. To include a random snippet
2665 of SQL verbatim, you specify it as a scalar reference. Consider this only
2666 as a last resort. Usually there is a better way. For example:
2669 priority => { '<', 2 },
2670 requestor => { -in => \'(SELECT name FROM hitmen)' },
2675 $stmt = "WHERE priority < ? AND requestor IN (SELECT name FROM hitmen)"
2678 Note that in this example, you only get one bind parameter back, since
2679 the verbatim SQL is passed as part of the statement.
2683 Never use untrusted input as a literal SQL argument - this is a massive
2684 security risk (there is no way to check literal snippets for SQL
2685 injections and other nastyness). If you need to deal with untrusted input
2686 use literal SQL with placeholders as described next.
2688 =head3 Literal SQL with placeholders and bind values (subqueries)
2690 If the literal SQL to be inserted has placeholders and bind values,
2691 use a reference to an arrayref (yes this is a double reference --
2692 not so common, but perfectly legal Perl). For example, to find a date
2693 in Postgres you can use something like this:
2696 date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
2701 $stmt = "WHERE ( date_column = date '2008-09-30' - ?::integer )"
2704 Note that you must pass the bind values in the same format as they are returned
2705 by L<where|/where(\%where, \@order)>. This means that if you set L</bindtype>
2706 to C<columns>, you must provide the bind values in the
2707 C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
2708 scalar value; most commonly the column name, but you can use any scalar value
2709 (including references and blessed references), L<SQL::Abstract> will simply
2710 pass it through intact. So if C<bindtype> is set to C<columns> the above
2711 example will look like:
2714 date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
2717 Literal SQL is especially useful for nesting parenthesized clauses in the
2718 main SQL query. Here is a first example :
2720 my ($sub_stmt, @sub_bind) = ("SELECT c1 FROM t1 WHERE c2 < ? AND c3 LIKE ?",
2724 bar => \["IN ($sub_stmt)" => @sub_bind],
2729 $stmt = "WHERE (foo = ? AND bar IN (SELECT c1 FROM t1
2730 WHERE c2 < ? AND c3 LIKE ?))";
2731 @bind = (1234, 100, "foo%");
2733 Other subquery operators, like for example C<"E<gt> ALL"> or C<"NOT IN">,
2734 are expressed in the same way. Of course the C<$sub_stmt> and
2735 its associated bind values can be generated through a former call
2738 my ($sub_stmt, @sub_bind)
2739 = $sql->select("t1", "c1", {c2 => {"<" => 100},
2740 c3 => {-like => "foo%"}});
2743 bar => \["> ALL ($sub_stmt)" => @sub_bind],
2746 In the examples above, the subquery was used as an operator on a column;
2747 but the same principle also applies for a clause within the main C<%where>
2748 hash, like an EXISTS subquery :
2750 my ($sub_stmt, @sub_bind)
2751 = $sql->select("t1", "*", {c1 => 1, c2 => \"> t0.c0"});
2752 my %where = ( -and => [
2754 \["EXISTS ($sub_stmt)" => @sub_bind],
2759 $stmt = "WHERE (foo = ? AND EXISTS (SELECT * FROM t1
2760 WHERE c1 = ? AND c2 > t0.c0))";
2764 Observe that the condition on C<c2> in the subquery refers to
2765 column C<t0.c0> of the main query : this is I<not> a bind
2766 value, so we have to express it through a scalar ref.
2767 Writing C<< c2 => {">" => "t0.c0"} >> would have generated
2768 C<< c2 > ? >> with bind value C<"t0.c0"> ... not exactly
2769 what we wanted here.
2771 Finally, here is an example where a subquery is used
2772 for expressing unary negation:
2774 my ($sub_stmt, @sub_bind)
2775 = $sql->where({age => [{"<" => 10}, {">" => 20}]});
2776 $sub_stmt =~ s/^ where //i; # don't want "WHERE" in the subclause
2778 lname => {like => '%son%'},
2779 \["NOT ($sub_stmt)" => @sub_bind],
2784 $stmt = "lname LIKE ? AND NOT ( age < ? OR age > ? )"
2785 @bind = ('%son%', 10, 20)
2787 =head3 Deprecated usage of Literal SQL
2789 Below are some examples of archaic use of literal SQL. It is shown only as
2790 reference for those who deal with legacy code. Each example has a much
2791 better, cleaner and safer alternative that users should opt for in new code.
2797 my %where = ( requestor => \'IS NOT NULL' )
2799 $stmt = "WHERE requestor IS NOT NULL"
2801 This used to be the way of generating NULL comparisons, before the handling
2802 of C<undef> got formalized. For new code please use the superior syntax as
2803 described in L</Tests for NULL values>.
2807 my %where = ( requestor => \'= submitter' )
2809 $stmt = "WHERE requestor = submitter"
2811 This used to be the only way to compare columns. Use the superior L</-ident>
2812 method for all new code. For example an identifier declared in such a way
2813 will be properly quoted if L</quote_char> is properly set, while the legacy
2814 form will remain as supplied.
2818 my %where = ( is_ready => \"", completed => { '>', '2012-12-21' } )
2820 $stmt = "WHERE completed > ? AND is_ready"
2821 @bind = ('2012-12-21')
2823 Using an empty string literal used to be the only way to express a boolean.
2824 For all new code please use the much more readable
2825 L<-bool|/Unary operators: bool> operator.
2831 These pages could go on for a while, since the nesting of the data
2832 structures this module can handle are pretty much unlimited (the
2833 module implements the C<WHERE> expansion as a recursive function
2834 internally). Your best bet is to "play around" with the module a
2835 little to see how the data structures behave, and choose the best
2836 format for your data based on that.
2838 And of course, all the values above will probably be replaced with
2839 variables gotten from forms or the command line. After all, if you
2840 knew everything ahead of time, you wouldn't have to worry about
2841 dynamically-generating SQL and could just hardwire it into your
2844 =head1 ORDER BY CLAUSES
2846 Some functions take an order by clause. This can either be a scalar (just a
2847 column name,) a hash of C<< { -desc => 'col' } >> or C<< { -asc => 'col' } >>,
2848 or an array of either of the two previous forms. Examples:
2850 Given | Will Generate
2851 ----------------------------------------------------------
2853 \'colA DESC' | ORDER BY colA DESC
2855 'colA' | ORDER BY colA
2857 [qw/colA colB/] | ORDER BY colA, colB
2859 {-asc => 'colA'} | ORDER BY colA ASC
2861 {-desc => 'colB'} | ORDER BY colB DESC
2863 ['colA', {-asc => 'colB'}] | ORDER BY colA, colB ASC
2865 { -asc => [qw/colA colB/] } | ORDER BY colA ASC, colB ASC
2868 { -asc => 'colA' }, | ORDER BY colA ASC, colB DESC,
2869 { -desc => [qw/colB/], | colC ASC, colD ASC
2870 { -asc => [qw/colC colD/],|
2872 ===========================================================
2876 =head1 SPECIAL OPERATORS
2878 my $sqlmaker = SQL::Abstract->new(special_ops => [
2882 my ($self, $field, $op, $arg) = @_;
2888 handler => 'method_name',
2892 A "special operator" is a SQL syntactic clause that can be
2893 applied to a field, instead of a usual binary operator.
2896 WHERE field IN (?, ?, ?)
2897 WHERE field BETWEEN ? AND ?
2898 WHERE MATCH(field) AGAINST (?, ?)
2900 Special operators IN and BETWEEN are fairly standard and therefore
2901 are builtin within C<SQL::Abstract> (as the overridable methods
2902 C<_where_field_IN> and C<_where_field_BETWEEN>). For other operators,
2903 like the MATCH .. AGAINST example above which is specific to MySQL,
2904 you can write your own operator handlers - supply a C<special_ops>
2905 argument to the C<new> method. That argument takes an arrayref of
2906 operator definitions; each operator definition is a hashref with two
2913 the regular expression to match the operator
2917 Either a coderef or a plain scalar method name. In both cases
2918 the expected return is C<< ($sql, @bind) >>.
2920 When supplied with a method name, it is simply called on the
2921 L<SQL::Abstract> object as:
2923 $self->$method_name ($field, $op, $arg)
2927 $field is the LHS of the operator
2928 $op is the part that matched the handler regex
2931 When supplied with a coderef, it is called as:
2933 $coderef->($self, $field, $op, $arg)
2938 For example, here is an implementation
2939 of the MATCH .. AGAINST syntax for MySQL
2941 my $sqlmaker = SQL::Abstract->new(special_ops => [
2943 # special op for MySql MATCH (field) AGAINST(word1, word2, ...)
2944 {regex => qr/^match$/i,
2946 my ($self, $field, $op, $arg) = @_;
2947 $arg = [$arg] if not ref $arg;
2948 my $label = $self->_quote($field);
2949 my ($placeholder) = $self->_convert('?');
2950 my $placeholders = join ", ", (($placeholder) x @$arg);
2951 my $sql = $self->_sqlcase('match') . " ($label) "
2952 . $self->_sqlcase('against') . " ($placeholders) ";
2953 my @bind = $self->_bindtype($field, @$arg);
2954 return ($sql, @bind);
2961 =head1 UNARY OPERATORS
2963 my $sqlmaker = SQL::Abstract->new(unary_ops => [
2967 my ($self, $op, $arg) = @_;
2973 handler => 'method_name',
2977 A "unary operator" is a SQL syntactic clause that can be
2978 applied to a field - the operator goes before the field
2980 You can write your own operator handlers - supply a C<unary_ops>
2981 argument to the C<new> method. That argument takes an arrayref of
2982 operator definitions; each operator definition is a hashref with two
2989 the regular expression to match the operator
2993 Either a coderef or a plain scalar method name. In both cases
2994 the expected return is C<< $sql >>.
2996 When supplied with a method name, it is simply called on the
2997 L<SQL::Abstract> object as:
2999 $self->$method_name ($op, $arg)
3003 $op is the part that matched the handler regex
3004 $arg is the RHS or argument of the operator
3006 When supplied with a coderef, it is called as:
3008 $coderef->($self, $op, $arg)
3016 Thanks to some benchmarking by Mark Stosberg, it turns out that
3017 this module is many orders of magnitude faster than using C<DBIx::Abstract>.
3018 I must admit this wasn't an intentional design issue, but it's a
3019 byproduct of the fact that you get to control your C<DBI> handles
3022 To maximize performance, use a code snippet like the following:
3024 # prepare a statement handle using the first row
3025 # and then reuse it for the rest of the rows
3027 for my $href (@array_of_hashrefs) {
3028 $stmt ||= $sql->insert('table', $href);
3029 $sth ||= $dbh->prepare($stmt);
3030 $sth->execute($sql->values($href));
3033 The reason this works is because the keys in your C<$href> are sorted
3034 internally by B<SQL::Abstract>. Thus, as long as your data retains
3035 the same structure, you only have to generate the SQL the first time
3036 around. On subsequent queries, simply use the C<values> function provided
3037 by this module to return your values in the correct order.
3039 However this depends on the values having the same type - if, for
3040 example, the values of a where clause may either have values
3041 (resulting in sql of the form C<column = ?> with a single bind
3042 value), or alternatively the values might be C<undef> (resulting in
3043 sql of the form C<column IS NULL> with no bind value) then the
3044 caching technique suggested will not work.
3048 If you use my C<CGI::FormBuilder> module at all, you'll hopefully
3049 really like this part (I do, at least). Building up a complex query
3050 can be as simple as the following:
3057 use CGI::FormBuilder;
3060 my $form = CGI::FormBuilder->new(...);
3061 my $sql = SQL::Abstract->new;
3063 if ($form->submitted) {
3064 my $field = $form->field;
3065 my $id = delete $field->{id};
3066 my($stmt, @bind) = $sql->update('table', $field, {id => $id});
3069 Of course, you would still have to connect using C<DBI> to run the
3070 query, but the point is that if you make your form look like your
3071 table, the actual query script can be extremely simplistic.
3073 If you're B<REALLY> lazy (I am), check out C<HTML::QuickTable> for
3074 a fast interface to returning and formatting data. I frequently
3075 use these three modules together to write complex database query
3076 apps in under 50 lines.
3078 =head1 HOW TO CONTRIBUTE
3080 Contributions are always welcome, in all usable forms (we especially
3081 welcome documentation improvements). The delivery methods include git-
3082 or unified-diff formatted patches, GitHub pull requests, or plain bug
3083 reports either via RT or the Mailing list. Contributors are generally
3084 granted full access to the official repository after their first several
3085 patches pass successful review.
3087 This project is maintained in a git repository. The code and related tools are
3088 accessible at the following locations:
3092 =item * Official repo: L<git://git.shadowcat.co.uk/dbsrgits/SQL-Abstract.git>
3094 =item * Official gitweb: L<http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=dbsrgits/SQL-Abstract.git>
3096 =item * GitHub mirror: L<https://github.com/dbsrgits/sql-abstract>
3098 =item * Authorized committers: L<ssh://dbsrgits@git.shadowcat.co.uk/SQL-Abstract.git>
3104 Version 1.50 was a major internal refactoring of C<SQL::Abstract>.
3105 Great care has been taken to preserve the I<published> behavior
3106 documented in previous versions in the 1.* family; however,
3107 some features that were previously undocumented, or behaved
3108 differently from the documentation, had to be changed in order
3109 to clarify the semantics. Hence, client code that was relying
3110 on some dark areas of C<SQL::Abstract> v1.*
3111 B<might behave differently> in v1.50.
3113 The main changes are :
3119 support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
3123 support for the { operator => \"..." } construct (to embed literal SQL)
3127 support for the { operator => \["...", @bind] } construct (to embed literal SQL with bind values)
3131 optional support for L<array datatypes|/"Inserting and Updating Arrays">
3135 defensive programming : check arguments
3139 fixed bug with global logic, which was previously implemented
3140 through global variables yielding side-effects. Prior versions would
3141 interpret C<< [ {cond1, cond2}, [cond3, cond4] ] >>
3142 as C<< "(cond1 AND cond2) OR (cond3 AND cond4)" >>.
3143 Now this is interpreted
3144 as C<< "(cond1 AND cond2) OR (cond3 OR cond4)" >>.
3149 fixed semantics of _bindtype on array args
3153 dropped the C<_anoncopy> of the %where tree. No longer necessary,
3154 we just avoid shifting arrays within that tree.
3158 dropped the C<_modlogic> function
3162 =head1 ACKNOWLEDGEMENTS
3164 There are a number of individuals that have really helped out with
3165 this module. Unfortunately, most of them submitted bugs via CPAN
3166 so I have no idea who they are! But the people I do know are:
3168 Ash Berlin (order_by hash term support)
3169 Matt Trout (DBIx::Class support)
3170 Mark Stosberg (benchmarking)
3171 Chas Owens (initial "IN" operator support)
3172 Philip Collins (per-field SQL functions)
3173 Eric Kolve (hashref "AND" support)
3174 Mike Fragassi (enhancements to "BETWEEN" and "LIKE")
3175 Dan Kubb (support for "quote_char" and "name_sep")
3176 Guillermo Roditi (patch to cleanup "IN" and "BETWEEN", fix and tests for _order_by)
3177 Laurent Dami (internal refactoring, extensible list of special operators, literal SQL)
3178 Norbert Buchmuller (support for literal SQL in hashpair, misc. fixes & tests)
3179 Peter Rabbitson (rewrite of SQLA::Test, misc. fixes & tests)
3180 Oliver Charles (support for "RETURNING" after "INSERT")
3186 L<DBIx::Class>, L<DBIx::Abstract>, L<CGI::FormBuilder>, L<HTML::QuickTable>.
3190 Copyright (c) 2001-2007 Nathan Wiger <nwiger@cpan.org>. All Rights Reserved.
3192 This module is actively maintained by Matt Trout <mst@shadowcatsystems.co.uk>
3194 For support, your best bet is to try the C<DBIx::Class> users mailing list.
3195 While not an official support venue, C<DBIx::Class> makes heavy use of
3196 C<SQL::Abstract>, and as such list members there are very familiar with
3197 how to create queries.
3201 This module is free software; you may copy this under the same
3202 terms as perl itself (either the GNU General Public License or
3203 the Artistic License)