Pointer magic

[fanf]

The following C type definition can be used for declaring local and global structure objects. You can initialize them as if they were bare structures, because C doesn't mind if you omit curly brackets in initializers (though gcc -Wall will complain). You can also use the typedef to declare function arguments, in which case the function will expect a pointer to the structure instead of a copy of it. Furthermore, when you use a variable declared with this typedef, it will be quietly converted into a pointer to the structure just as is expected by the function. This avoids a load of & operators and gives you a sort of poor-man's C++ pass-by-reference.

        typedef struct mytype {
                /* member declarations */
        } mytype[1];

        mytype var;

        int func(mytype arg);

        func(var);

ETA: it seems this trick is used by GMP (see the last paragraph of that page)

[Poll #1092168]

Comments

[simont]

This is the same trick used by jmp_buf, isn't it? Hence why you don't have to specify the & when passing a jmp_buf to a function which is obviously going to modify it (such as setjmp).

[ewx.livejournal.com]

I use it often and like it because I can always use -> and leave out the & regardless of whether I'm in the function that declared the auto object or one that has been passed a pointer to it; as well as not having to care where I am when writing, it means search/search and replace is simplified. An interesting example of typing more characters in an essentially mechanical way nonetheless reducing the total amount of work.

[fanf]

So it is :-) Though it seems older unices just used an array of ints.

[compilerbitch.livejournal.com]

Just use C++. You know you want to. :-P

(Brought to you by the department of trolling comp.lang.c)

[compilerbitch.livejournal.com]

PS: Yes, I ticked all of the boxes. So sue me already. :)

[pm215]

I don't like this for the same reason I don't like C++'s pass-by-reference -- it hides the distinction between passing a thing and passing a pointer to a thing, which IMHO is a fundamental difference that you really want to know about.

[bens_dad]

Hmm. My C appears to be too rusty to get the full subtlety of this but I think I agree with pm215 that in C you need to know whether you have a thing or a pointer to a thing. Unless you can use this consistently and everywhere, so that you never need to remember which, it is probably more dangerous to forget that you need to remember.

IIRC K&R C didn't allow you to pass arrays, so I'd be very uncomfortable about using this on a pre-ANSI C compiler.
(That isn't likely these days, so that is really an aesthetic complaint.)

[simont]

I agree with this.

Also if you want to define a more complex type with one of these things as its base type, it's hard to work out how. When I was working with GMP a while back I occasionally wanted to write a function which took a pointer to an array of mpf_t; I managed it in the end by just adding or removing *s until the compiler stopped complaining, but I wouldn't say I understand the syntactic rules which make the right answer right and the wrong ones wrong. (And, just to put that in perspective, I do feel that I understand pretty much all the rest of the C declarator syntax, which I know isn't a statement all C programmers could make!)

[gareth-rees.livejournal.com]

If in doubt, I suggest using something like this:

typedef mpf_t array_of_unknown_size_of_mpf_t[];
typedef array_of_unknown_size_of_mpf_t *pointer_to_array_of_unknown_size_of_mpf_t;

[gerald_duck]

I don't like some of the darker corners of C's array behaviour anyway, and this exhibits them in a place where you're not expecting them. Particularly, while func(var) passes by reference, mytype another = var; makes a copy. C++'s references have their faults, but argument binding behaviour differing from assignment behaviour isn't one of them.

[ewx.livejournal.com]

I think you've confused it with the old struct passing behavior; AFAIK the array-to-pointer decay has always been that way.

[ewx.livejournal.com]

Oh, and all sorts of standard modern things don't work on pre-ANSI implementations: const and prototypes for instance. So suggesting that as even just an aesthetic reason is frankly rather silly.

[pm215]

Yeah, even gcc requires an ANSI C89 compiler these days. K&R non-prototyped function syntax is almost as dead as the "var =+ 4;" autoincrement syntax. (Aside: I wonder if the Green Hills C compiler still has the option switch to support =+ ...)

[bens_dad]

I was indeed thinking of that.

[drj11.livejournal.com]

mytype another = var isn't valid C, so it doesn't make a copy.

I've seen this trick used for va_list so that you could pass vararg lists to subroutines and have it work correctly, though the general form of this is not required by the standard.

I wouldn't really consider this a dark corner of C's array behaviour, it's simply one of those things you have to know before calling yourself a C programmer.