else {
require mro;
}
+
+ *SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION = $ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}
+ ? sub () { 0 }
+ : sub () { 1 }
+ ;
}
#======================================================================
# intersted in are much more limited than the fullblown thing, and
# this is a very hot piece of code
(
- # FIXME - DBI needs fixing to stringify regardless of DBD
- #
# simply using ->can('(""') can leave behind stub methods that
# break actually using the overload later (see L<perldiag/Stub
# found while resolving method "%s" overloading "%s" in package
# either has stringification which DBI SHOULD prefer out of the box
grep { *{ (qq[${_}::(""]) }{CODE} } @{ $_[2] = mro::get_linear_isa( $_[1] ) }
or
- # has nummification and fallback is *not* disabled
+ # has nummification or boolification, AND fallback is *not* disabled
(
- grep { *{"${_}::(0+"}{CODE} } @{ mro::get_linear_isa( $_[1] ) }
+ SQL::Abstract::_ENV_::DETECT_AUTOGENERATED_STRINGIFICATION
+ and
+ (
+ grep { *{"${_}::(0+"}{CODE} } @{$_[2]}
+ or
+ grep { *{"${_}::(bool"}{CODE} } @{$_[2]}
+ )
and
(
# no fallback specified at all
my ($sql, @bind) = $self->$method($where, $logic);
- # DBIx::Class directly calls _recurse_where in scalar context, so
- # we must implement it, even if not in the official API
- return wantarray ? ($sql, @bind) : $sql;
+ # DBIx::Class used to call _recurse_where in scalar context
+ # something else might too...
+ if (wantarray) {
+ return ($sql, @bind);
+ }
+ else {
+ belch "Calling _recurse_where in scalar context is deprecated and will go away before 2.0";
+ return $sql;
+ }
}
my %data = (
name => 'Bill',
- date_entered => \["to_date(?,'MM/DD/YYYY')", "03/02/2003"],
+ date_entered => \[ "to_date(?,'MM/DD/YYYY')", "03/02/2003" ],
);
The first value in the array is the actual SQL. Any other values are
=head1 METHODS
-The methods are simple. There's one for each major SQL operation,
+The methods are simple. There's one for every major SQL operation,
and a constructor you use first. The arguments are specified in a
-similar order to each method (table, then fields, then a where
+similar order for each method (table, then fields, then a where
clause) to try and simplify things.
=head2 new(option => 'value')
sub called C<bind_fields()> or something and reuse it repeatedly. You still
get a layer of abstraction over manual SQL specification.
-Note that if you set L</bindtype> to C<columns>, the C<\[$sql, @bind]>
+Note that if you set L</bindtype> to C<columns>, the C<\[ $sql, @bind ]>
construct (see L</Literal SQL with placeholders and bind values (subqueries)>)
will expect the bind values in this format.
On failure returns C<undef>, on sucess returns a B<scalar> reference
to the original supplied argument.
+=over
+
+=item * Note
+
+The stringification overloading detection is rather advanced: it takes
+into consideration not only the presence of a C<""> overload, but if that
+fails also checks for enabled
+L<autogenerated versions of C<"">|overload/Magic Autogeneration>, based
+on either C<0+> or C<bool>.
+
+Unfortunately testing in the field indicates that this
+detection B<< may tickle a latent bug in perl versions before 5.018 >>,
+but only when very large numbers of stringifying objects are involved.
+At the time of writing ( Sep 2014 ) there is no clear explanation of
+the direct cause, nor is there a manageably small test case that reliably
+reproduces the problem.
+
+If you encounter any of the following exceptions in B<random places within
+your application stack> - this module may be to blame:
+
+ Operation "ne": no method found,
+ left argument in overloaded package <something>,
+ right argument in overloaded package <something>
+
+or perhaps even
+
+ Stub found while resolving method "???" overloading """" in package <something>
+
+If you fall victim to the above - please attempt to reduce the problem
+to something that could be sent to the L<SQL::Abstract developers
+|DBIx::Class/GETTING HELP/SUPPORT>
+(either publicly or privately). As a workaround in the meantime you can
+set C<$ENV{SQLA_ISVALUE_IGNORE_AUTOGENERATED_STRINGIFICATION}> to a true
+value, which will most likely eliminate your problem (at the expense of
+not being able to properly detect exotic forms of stringification).
+
+This notice and environment variable will be removed in a future version,
+as soon as the underlying problem is found and a reliable workaround is
+devised.
+
+=back
+
=head2 is_literal_value
Determines if the supplied argument is a literal value as understood by this
@bind = ('2', '5', 'nwiger');
If you want to include literal SQL (with or without bind values), just use a
-scalar reference or array reference as the value:
+scalar reference or reference to an arrayref as the value:
my %where = (
date_entered => { '>' => \["to_date(?, 'MM/DD/YYYY')", "11/26/2008"] },
Which would generate:
- $stmt = "WHERE date_entered > "to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
+ $stmt = "WHERE date_entered > to_date(?, 'MM/DD/YYYY') AND date_expires < now()";
@bind = ('11/26/2008');
Because, in Perl you I<can't> do this:
- priority => { '!=', 2, '!=', 1 }
+ priority => { '!=' => 2, '!=' => 1 }
As the second C<!=> key will obliterate the first. The solution
is to use the special C<-modifier> form inside an arrayref:
That would yield:
- WHERE ( user = ? AND (
- ( workhrs > ? AND geo = ? )
- OR ( workhrs < ? OR geo = ? )
- ) )
+ $stmt = "WHERE ( user = ?
+ AND ( ( workhrs > ? AND geo = ? )
+ OR ( workhrs < ? OR geo = ? ) ) )";
+ @bind = ('nwiger', '20', 'ASIA', '50', 'EURO');
=head3 Algebraic inconsistency, for historical reasons
in Postgres you can use something like this:
my %where = (
- date_column => \[q/= date '2008-09-30' - ?::integer/, 10/]
+ date_column => \[ "= date '2008-09-30' - ?::integer", 10 ]
)
This would create:
@bind = ('10');
Note that you must pass the bind values in the same format as they are returned
-by L</where>. That means that if you set L</bindtype> to C<columns>, you must
-provide the bind values in the C<< [ column_meta => value ] >> format, where
-C<column_meta> is an opaque scalar value; most commonly the column name, but
-you can use any scalar value (including references and blessed references),
-L<SQL::Abstract> will simply pass it through intact. So if C<bindtype> is set
-to C<columns> the above example will look like:
+by L<where|/where(\%where, \@order)>. This means that if you set L</bindtype>
+to C<columns>, you must provide the bind values in the
+C<< [ column_meta => value ] >> format, where C<column_meta> is an opaque
+scalar value; most commonly the column name, but you can use any scalar value
+(including references and blessed references), L<SQL::Abstract> will simply
+pass it through intact. So if C<bindtype> is set to C<columns> the above
+example will look like:
my %where = (
- date_column => \[q/= date '2008-09-30' - ?::integer/, [ dummy => 10 ]/]
+ date_column => \[ "= date '2008-09-30' - ?::integer", [ {} => 10 ] ]
)
Literal SQL is especially useful for nesting parenthesized clauses in the
the expected return is C<< ($sql, @bind) >>.
When supplied with a method name, it is simply called on the
-L<SQL::Abstract/> object as:
+L<SQL::Abstract> object as:
$self->$method_name ($field, $op, $arg)
Where:
- $op is the part that matched the handler regex
$field is the LHS of the operator
+ $op is the part that matched the handler regex
$arg is the RHS
When supplied with a coderef, it is called as:
the expected return is C<< $sql >>.
When supplied with a method name, it is simply called on the
-L<SQL::Abstract/> object as:
+L<SQL::Abstract> object as:
$self->$method_name ($op, $arg)
=item *
-support for literal SQL through the C<< \ [$sql, bind] >> syntax.
+support for literal SQL through the C<< \ [ $sql, @bind ] >> syntax.
=item *
projects / dbsrgits/SQL-Abstract.git / blobdiff