Re: Passing-by-reference

On Thu, 23 Nov 2000, Andi Gutmans wrote: > At 03:11 PM 11/23/00 -0500, Ignacio Vazquez-Abrams wrote: > > > >c) Throw an error. More than likely this will indicate a coding problem, and > >just passing a copy will cause a developer to scratch her head and say > >"WTF?" for several months while trying to figure out the problem(s). Leaving > >behavior as undefined will just cause memory corruption problems for sloppy > >developers, which is what we want to get away from in the first place. > > Throwing an error is a problem because we can't really detect it. With my > patch the undefined behavior will not cause memory corruption anymore but > will either throw an error or continue working (i.e. undefined). > In the manual it would say something like: > Receiving arguments by reference: > ---------------------------------------- > Three ways of passing an argument to a function which wants to receive its > argument by reference: > a) foo($a) // Passing a variable > b) foo(new foo()) // Passing a newly created object > c) foo(MyFactory()) // Where MyFactory() is explicitly defined to return > the return value by reference, which is done by defining it as "function > &MyFactor(){...}" > > ALL OTHER ARGUMENT PASSING RESULTS IN UNDEFINED BEHAVIOR! > > How's that? :) > Andi > > Andi > --- > Andi Gutmans <andi@zend.com> > http://www.zend.com/ > Hmm. I guess the only way to detect something like that would be to have another "unreferrable" field on a value and set it when the value is a constant expression (e.g., "Hello, world", 5+4, a value returned from a non-reference-returning function, etc.), and clear it when assigned to a variable. A feature for 5.0 perhaps? -- Ignacio Vazquez-Abrams <ignacio@openservices.net>