Casting 0 to `extern "C" fn` is permitted

[MicahChalmer]

Now that #8666 is in place, there is not a good way to make a NULL extern "C" fn. You can do it, but you have to specify the whole type of the function pointer again with as:

0 as extern "C" fn(args...) -> ReturnType

using std::ptr::null() will not compile, and gives:

hello_nullfuncptrs.rs:4:8: 4:25 error: mismatched types: expected `extern "C" fn(int, int) -> int` but found `*<V0>` (expected extern fn but found *-ptr)

Some C APIs really do need to have NULL passed in for function pointers. Right now, using as like the code above is the only way to create such a null pointer; For those cases where you need to pass a NULL into a C API, it would be nice if std::ptr::null() would work for that, without having to repeat the whole type (including all arguments and return type) of the function pointer in order to do it.

UPDATE: Title amended by @nikomatsakis to reflect the real issue at hand.

[MicahChalmer]

This was previously both about this and #8728, but edited to be about the inference issue only; #8728 is for the safety hole.

[nikomatsakis]

The correct type for a nullable extern pointer is Option<extern "C" fn>, and you can pass null with None.

[nikomatsakis]

In light of that, I'm not sure if there is still a bug here or not?

[MicahChalmer]

Ah! That does work, and makes sense. I didn't know you could use Option that way and it would still match the C signature of just a pointer--Option is an enum after all, and the current version of the FFI reference specifically says there are no guarantees about the layout of an enum. But as far as usability goes it's great.

All I would ask is for this to be documented the FFI tutorial and in the reference manual. I wouldn't have expected it to be there yet, since nothing from #8666 is documented yet, which I assume is a different issue altogether, or at least in progress.

(Given this, am I expected to close the bug here? Or should only people with commit access be doing that?)

[nikomatsakis]

Do not close the bug, 0 as extern "C" fn should not type check. I will edit the title and so forth. And I agree that this use of Option should be documented, both with respect to the fact that the enum layout in this case is guaranteed and to give examples of how it is done.

[nikomatsakis]

oh, nm, dup of #8728

[olson-dan]

The correct type for a nullable extern pointer is Option<extern "C" fn>, and you can pass null with None.

How does one deal with this in the case of APIs that are not correctly wrapped in an option? Like this one:
https://docs.rs/libz-sys/1.0.20/libz_sys/struct.z_stream.html

Edit:
I found that std::mem:zeroed() works, so I guess that this is an ok workaround.

[BatmanAoD]

@nikomatsakis Is it still true that None should automatically convert to null when passed to an external function, or did that change? It doesn't seem to work for me, so I don't know if that feature was removed before 1.0 or if I'm misunderstanding how to use it.