Windows: sa_family_t should use ADDRESS_FAMILY from windows-sys

[nerditation]

in #414 socklen_t and sa_family_t were changed to use types from windows-sys, which themselves were added to the win32metadata by request, in microsoft/win32metadata#1041 (both as NativeTypedefs). however, sa_family_t was later changed to MetadataTypedef in microsoft/win32metadata#1538 because it was an "invented" handle type (meaning it doesn't exist in any of the Windows SDK header files but only in win32metadata). socklen_t, on the other hand, does exists in the Windows SDK header "WS2tcpip.h", so it remains NativeTypedef.

it turns out that the equivalent to POSIX sa_family_t in Winsock2 is the ADDRESS_FAMILY type, and it existed forever. I think it's a mistake to request a win32metadata only typedef while ingoring the proper Windows SDK type,- and I'd suggest we change our sa_family_t to use the proper type, i.e. windows_sys::Win32::Networking::Winsock::ADDRESS_FAMILY, which also makes it consistent with other Winsock2 APIs, for instance, in windows-sys (and also windows), the type of SOCKADDR::sa_family and SOCKADDR_STORAGE::sa_family is ADDRESS_FAMILY, the constants AF_INET, AF_INET6 are defined using type ADDRESS_FAMILY, and some iphelper functions like GetIpForwardTable2() accept ADDRESS_FAMILY as argument.

example code in the `windows` crate where `ADDRESS_FAMILY` is used

SOCKADDR

https://github.com/microsoft/windows-rs/blob/891c469706d314fcf0e3a7913f45970ce11b13f0/crates/libs/windows/src/Windows/Win32/Networking/WinSock/mod.rs#L13218-L13223

SOCKADDR_IN::sin_family

https://github.com/microsoft/windows-rs/blob/891c469706d314fcf0e3a7913f45970ce11b13f0/crates/libs/windows/src/Windows/Win32/Networking/WinSock/mod.rs#L13314-L13321

AF_INET and AF_INET6

https://github.com/microsoft/windows-rs/blob/891c469706d314fcf0e3a7913f45970ce11b13f0/crates/libs/windows/src/Windows/Win32/Networking/WinSock/mod.rs#L1782-L1785

btw, I discovered #414 when I was investigating the discrepancy of wrapper types from the windows crate. microsoft/windows-rs#2627

[Thomasdezeeuw]

That seems reasonable.

One possible difficulty is that SockAddr::family returns sa_family_t exported by the sys module, currently that is windows_sys::Win32::Networking::WinSock::sa_family_t for Windows. It could be a breaking change to change it, even though it's all the same primitive type underneath.

[nerditation]

It could be a breaking change to change it, even though it's all the same primitive type underneath.

it's a valid concern, but I think changing from sa_family_t to ADDRESS_FAMILY is no different than changing from u16 to sa_family_t as done in #414

what's more, how is SockAddr::family() supposed to be used by the caller? they compare the return value to AF_INET, AF_INET6, right? in windows-sys these constants are defined as:

pub const AF_INET: ADDRESS_FAMILY = 2u16;
pub const AF_INET6: ADDRESS_FAMILY = 23u16;

so in this regard, the caller code like this

let family = sockaddr.family();
if family == AF_INET {
    todo!()
}

is technically already "broken" if we keep it as the current definition, and change the return type to ADDRESS_FAMILY is no more "broken" than it's current form (but actually more "correct").

[Thomasdezeeuw]

it's a valid concern, but I think changing from sa_family_t to ADDRESS_FAMILY is no different than changing from u16 to sa_family_t as done in #414

Fair point.

what's more, how is SockAddr::family() supposed to be used by the caller? they compare the return value to AF_INET, AF_INET6, right? in windows-sys these constants are defined as:

pub const AF_INET: ADDRESS_FAMILY = 2u16;
pub const AF_INET6: ADDRESS_FAMILY = 23u16;

To be honest, I've only done this on Unix, so I don't have a good answer to this.

so in this regard, the caller code like this

let family = sockaddr.family();
if family == AF_INET {
    todo!()
}

is technically already "broken" if we keep it as the current definition, and change the return type to ADDRESS_FAMILY is no more "broken" than it's current form (but actually more "correct").

Yeah, fair enough. Do you want to send a pr to change the type?

[nerditation]

Do you want to send a pr to change the type?

sure, #463 is created

[Thomasdezeeuw]

This was done in #463.