empty structs

[bvdberg]

Currently C2 allows empty structs:

type Foo struct {}

I have found no use of this however, so then i would like to forbid it (since it makes life easier for auto-conversions).
Currently there are some empty structs in library headers, defining empty, opaque structs.
Does anyone else have a use for this?

[lerno]

I removed it from C3 because it creates very weird corner cases. While it also solves some corner cases, the disadvantages outweigh the advantages in C3. This also means there are no zero sized arrays.

[bvdberg]

C2 allows zero size arrays inside struct definitions, because there is a use-case for that (tail-allocated stuff). Otherwise zero-size arrays are also not allowed..

Ok, I'll put disallowing empty structs on my list.. thx

[lerno]

I differentiate between flexible array members (allowed) and zero sized arrays (not allowed)

[chqrlie]

The question should not be should we allow empty structs for which we have not found a use case?, it should be do empty structs pose a problem? ? If they do, then a good fix for the problem is to disallow them.

So far the only problem I see with empty structs is C does not allow them, so extra code would be required in the C back-end to generate compilable code.

A potential use case is compile time extendable structs and unions: just like you can define extendable enums and arrays, using [+] and += to gather the elements across source files, one could imagine extendable types where extra members could be added in different places. This could make sense for unions of pointer types. I am not sure what syntax would be elegant for this, but starting from an empty struct or union might make sense.

Similarly, empty enum are disallowed, yet would make sense if they get extended later.

[lerno]

Here are a few special things about zero sized structs:

ZeroStruct x;
ZeroStruct* y = &x; // What does this point to?
ZeroStruct[4] z;
z[2]; // What does this do?
&z[2] - &z[0]; // What's the diff?

[lerno]

C++ requires each element to have a unique address, so a zero sized struct is still 1 byte.

[chqrlie]

I guess a struct without any members could have one padding byte, making its length 1, which makes your examples easy to solve:

type ZeroStruct struct {}  // an empty struct type for the sake of discussion until proven useful 
ZeroStruct x;  // a one byte object without any accessible fields
ZeroStruct* y = &x; // `y` points to the memory storage for `x`
ZeroStruct[4] z;  // `z` has a size of 4 bytes
z[2]; // This is an error: an expression without effect
usize diff = &z[2] - &z[0]; // this evaluates to 2

[lerno]

Yes, but note that the GCC extension of zero size structs for C, leaves the struct at zero size.

[bvdberg]

I would rather not have to work around the issue of zero-sized structs. I'm always in favor of keeping things as simple as possible; there are plenty of other features C2 developers can shoot themselves in the foot with...