Usage of struct tag results in odd translation about ctod

[dkorpel]

Fixed by 7cd36c1

I now assume that bodyless enum/struct types are already defined, even when they have pointer types.

[dkorpel]

It does that because in C you can declare an opaque struct in-line:

void f(struct S *s)
{

}

However, in your examples, it should deduce that there is no pointer type, so it has to be defined already and should not emit a type definition.

[schveiguy]

The example function don't match the actual code, where it is a pointer. But the struct is defined (possibly in another module), so redefining the struct is still wrong.

For sure, redefining the struct inside a nested struct is wrong.

However, this is D. Can you not record which structs may or may not have been defined, and just spit out some struct S at the end if you hadn't seen it yet?

[schveiguy]

An option to disable this "struct prototype" feature would probably work too. Most code is going to be consistent in how they use structs.

[schveiguy]

Just tested by compiling, and it does indeed complain if you redefine the struct within the same module, even to just define as a prototype.

[dkorpel]

I suppose a future enhancement would be to manage a symbol table to know which symbols have been defined already, but unless you also read all #includes you still cannot be sure.

[schveiguy]

Yeah, it's probably not a huge deal, because defining the struct in the module is such a different thing than defining it again in C.