-Wold-style-declaration: Add diagnostic

[alejandro-colomar]

Hi!

GCC has this diagnostic for obsolescent uses of storage class specifiers not the first specifier in a declaration.

typedef int ti;
int typedef it;

static int si;
int static is;

alx@debian:~/tmp$ gcc -S -Wold-style-declaration reversed.c
reversed.c:2:1: warning: ‘typedef’ is not at beginning of declaration [-Wold-style-declaration]
    2 | int typedef it;
      | ^~~
reversed.c:5:1: warning: ‘static’ is not at beginning of declaration [-Wold-style-declaration]
    5 | int static is;
      | ^~~
alx@debian:~/tmp$ clang -S -Weverything reversed.c -Wno-unused
alx@debian:~/tmp$ 

Would you mind adding this diagnostic?

[frederick-vs-ja]

The deprecation has been present since C89.

The placement of a storage-class specifier other than at the beginning of the declaration specifiers in a declaration is an obsolescent feature.

But I haven't found the reason, and there doesn't seem any follow-up paper discussing this. It's also weird that such use isn't deprecated/obsolescent in C++.

[alejandro-colomar]

The deprecation has been present since C89.

The placement of a storage-class specifier other than at the beginning of the declaration specifiers in a declaration is an obsolescent feature.

But I haven't found the reason, and there doesn't seem any follow-up paper discussing this. It's also weird that such use isn't deprecated/obsolescent in C++.

We mentioned the idea of tightening C to disallow that recently in WG14. There hasn't been any paper, but I could certainly write one. It would be interesting to have GCC and Clang agree on this and have a diagnostic; possibly enabled by default.

[alejandro-colomar]

The deprecation has been present since C89.

The placement of a storage-class specifier other than at the beginning of the declaration specifiers in a declaration is an obsolescent feature.

But I haven't found the reason, and there doesn't seem any follow-up paper discussing this. It's also weird that such use isn't deprecated/obsolescent in C++.

We mentioned the idea of tightening C to disallow that recently in WG14. There hasn't been any paper, but I could certainly write one. It would be interesting to have GCC and Clang agree on this and have a diagnostic; possibly enabled by default.

There seems to be a use case (within glibc internals, which are quite weird), where static not being first makes sense, through macro expansion.

I think it could be done in a different way, but it works for glibc, so I guess we won't do that in the standard. However, a -Wextra diagnostic still makes sense, since glibc is very weird doing that, and most people would benefit from such a diagnostic.