Comments on: NULL as a function argument should be illegal

By: Colin Barrett

Colin Barrett — Sat, 10 Mar 2007 01:33:04 +0000

assert()ing on illegal pre-conditions is extremely useful (although you need to get used to looking at the top of the method for them ;). I'm not sure how asserting on post-conditions would be useful within a method: shouldn't you assert when you find an inconsistency at any point, not just post-flight?

By: David Smith

David Smith — Tue, 27 Feb 2007 16:54:20 +0000

So on a related note, what do you feel about assert()ing pre- and post-conditions for methods? Obviously the technique is a lot more powerful in languages like Eiffel where it can be used at compile time to disallow broken programs, but it still seems potentially useful in other languages.

By: Peter Hosey

Peter Hosey — Mon, 26 Feb 2007 18:36:39 +0000

Phil: Yup, I totally agree—for an output (return-something-by-reference) argument, using NULL to mean “no thanks” should always be allowed.

I'll update the post later today, but I'll point out now that this really only applies to C and its descendants. Other programming languages—those with a first-class sequence type—should simply return everything in a tuple/list/array and let the caller unpack the values and throw away what he doesn't want. (Fictional Python example: x, y = myView.frameOrigin())

By: Phil

Phil — Mon, 26 Feb 2007 18:26:59 +0000

I think you miss an important use of NULL, and one i believe to be valid. (example from NSXMLNode)

-(NSArray *)nodesForXPath:(NSString *)xpath error:(NSError **)error

When your methods are providing a way to return additional data into a user-specified memory location, being able to optionally say "no thanks, I don't want that data" is useful, especially when following a standard model (like the NSError one). When it comes to using NULL/nil as a data providing argument, I think you're right, however.

By: David Smith

David Smith — Sun, 25 Feb 2007 21:36:44 +0000

AIContactController is (or was, before I changed some of it) a particularly flagrant abuser of this principle. It definitely tends to conceal bugs, and imo NULL arguments should probably assert.

By: Evan

Evan — Sun, 25 Feb 2007 16:39:07 +0000

I've definitely made this mistake with NSNotificationCenter before -- observing for an object I "know" is non-NULL, then being confused when I receive notifications for everything because a buglet elsewhere had nulled the variable. Making it illegal is a great design idea.