Bill Wood wrote:

Could you briefly elaborate on what you mean by "hybrid variables"?

According to Google, hybrid in genetics means "The offspring of genetically dissimilar parents or stock, especially the offspring produced by breeding plants or animals of different varieties, species, or races." It's kind of like that - but for variables. The typical example in C is: mem = malloc(1024) Malloc returns 0 to indicate that memory cannot be allocated, or a memory address if it can. The variable mem is a so-called hybrid variable; it crunches together 2 different concepts: a boolean value (could I allocate memory?) and an address value (what is the address where I can find my allocated memory). It's considered a bad idea because it makes it easy for programmers to use the value inappropriately - witness the number of programmers who pass in 0 as a memory location. The suggested solution is to give each variable one meaning. So, I could define, say, a variable called success, and do something like mem =improved_ malloc(1024, & success) if(success) { do_something(mem) } else { oops() } Admittedly, there's nothing to force the programmer to actually check the success variable; but at least the idea is that everything is much more explicit, and the programmer is less likely to shoot himself in the foot. ___________________________________________________________ Yahoo! Messenger - NEW crystal clear PC to PC calling worldwide with voicemail http://uk.messenger.yahoo.com

Lennart Augustsson wrote:

Mark Carter wrote:

...

The typical example in C is: mem = malloc(1024) Malloc returns 0 to indicate that memory cannot be allocated, or a memory address if it can. The variable mem is a so-called hybrid variable; it crunches together 2 different concepts: a boolean value (could I allocate memory?) and an address value (what is the address where I can find my allocated memory).

It's considered a bad idea because it makes it easy for programmers to use the value inappropriately - witness the number of programmers who pass in 0 as a memory location.

This is a bad idea in C, because you cannot force programmers to test the return value properly.

The Maybe type in Haskell is a good idea, because you must test the a Maybe value to extract the real value. (Using the Maybe monad this can be hidden.)

Yes, I didn't mean to imply that Haskell had got its ideas fundamentally wrong. I think that the main thrust of the argument is that in something like C, hybrid variables make it easy for a programmer to inadvertently use them wrongly, as opposed to just be lazy. I suppose there's some element of debate on the matter: for example, what C programmer doesn't honestly know that memory allocation can fail. And for an unfamiliar function, why doesn't the programmer read the documentation to find out what constitutes valid and invalid return values. OTOH, I think Charles Moore is quoted as saying that if he want to add 1 to the letter A, then he didn't want the programming language to stop him. Which is quite a contrast to Haskell and its notions on safety. I'm not flamebaiting, you understand, I'm just pointing out the various viewpoints to the argument. ___________________________________________________________ How much free photo storage do you get? Store your holiday snaps for FREE with Yahoo! Photos http://uk.photos.yahoo.com

Compare: int *p=...; int x=*p; and: let p = ... Just x = p So actually, there is few difference between dereferencing a pointer without checking for 0, and extracting the Maybe value without handling Nothing, apart from that it leads to undefined behavior in C which in fact isn't really a point against "hybrid variables". On 9/20/05, Lennart Augustsson wrote:

Mark Carter wrote:

...

The typical example in C is: mem = malloc(1024) Malloc returns 0 to indicate that memory cannot be allocated, or a memory address if it can. The variable mem is a so-called hybrid variable; it crunches together 2 different concepts: a boolean value (could I allocate memory?) and an address value (what is the address where I can find my allocated memory).

It's considered a bad idea because it makes it easy for programmers to use the value inappropriately - witness the number of programmers who pass in 0 as a memory location.

This is a bad idea in C, because you cannot force programmers to test the return value properly.

The Maybe type in Haskell is a good idea, because you must test the a Maybe value to extract the real value. (Using the Maybe monad this can be hidden.)

-- Lennart

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

. . .

The typical example in C is: mem = malloc(1024) Malloc returns 0 to indicate that memory cannot be allocated, or a memory address if it can. The variable mem is a so-called hybrid variable; it crunches together 2 different concepts: a boolean value (could I allocate memory?) and an address value (what is the address where I can find my allocated memory).

An infamous example would be the convention in Common Lisp that nil, the empty list, is also "false" for conditionals while anything else is "true" for conditionals. So the member function ('a * ['a] -> ['a]) can be used either as a predicate or a function returning a useful value.

It's considered a bad idea because it makes it easy for programmers to use the value inappropriately - witness the number of programmers who pass in 0 as a memory location. The suggested solution is to give each

And the Scheme community chose "#f" and "#t" for boolean values so you had to be a little more explicit about what you were doing. I mostly agree with the "tightening-up", but there are times when I really miss the nil hacks :-) -- Bill Wood bill.wood@acm.org

Mark Carter wrote:

...

Could you briefly elaborate on what you mean by "hybrid variables"?

According to Google, hybrid in genetics means "The offspring of genetically dissimilar parents or stock, especially the offspring produced by breeding plants or animals of different varieties, species, or races." It's kind of like that - but for variables.

The typical example in C is: mem = malloc(1024) Malloc returns 0 to indicate that memory cannot be allocated, or a memory address if it can. The variable mem is a so-called hybrid variable; it crunches together 2 different concepts: a boolean value (could I allocate memory?) and an address value (what is the address where I can find my allocated memory).

Well in that case, Maybe provides the perfect example of how to implement hybrid variables correctly. The types "Ptr a" and "Maybe (Ptr a)" are distinct. If you try to pass the latter to a function which expects the former, you'll get a compile-time error. You first have to extract the underlying value, which means that you need to match against (Just x). If the wrapped value is Nothing, you'll get an exception. Furthermore, if you forget to handle the Nothing case, you'll get a compile-time warning. In C, there's no way to distinguish (using the type system) between a possibly-null pointer and a non-null pointer. Using a pair of a boolean and a pointer is the wrong approach because the pointer is meaningless if the boolean is false, but the type system won't prevent you from using the value of the pointer in that case. A more general example is structures where certain fields are only valid in certain circumstances (e.g. depending upon the "type" field). Haskell-style sum types, (of which Maybe is an example) are a much better solution, as the the fields only exist when they are meaningful. -- Glynn Clements