x/sys/unix: add support for vectorized ("scatter/gather") IO syscalls on Darwin

[mschoenlaub]

Proposal Details

The following syscalls could be supported on modern OSX versions (Darwin 11 and newer)

• unix.Readv(fd int, iovs [][]byte) (n int, err error)
• unix.Preadv(fd int, iovs [][]byte, offset int64) (n int, err error)
• unix.Writev(fd int, iovs [][]byte) (n int, err error)
• unix.Pwritev(fd int, iovs [][]byte, offset int64) (n int, err error)

I think https://go-review.googlesource.com/c/sys/+/548795 contains everything needed to implement this on Darwin.
In cases where the version is too old, the public functions in unix would simply return ENOSYS instead of running into a dyld error due to trying to jump to a function that doesn't exist.

[mauri870]

cc @golang/darwin

[gopherbot]

Change https://go.dev/cl/548795 mentions this issue: unix: support Readv, Preadv, Writev and Pwritev for darwin

[adonovan]

Why just Darwin? p{read,write}v have been supported on Linux and various BSDs for a while. Ideally the sys interface would have implementations for all platforms that support it, even if the macOS implementation needs an additional compatibility check.

[Update: I now see that we have a Linux implementation already. But my point is that the proposal should cover the API change, and implementations for all platforms can follow at their own pace if they conform to the approved proposal.]

[mschoenlaub]

Hm, yeah, @adonovan I think you are right. The goal should be to implement it on all platforms supported, but I agree on these implementations better coming in at their own pace.
However, I think I wasn't fully aware of actually proposing an API change of any sort, because due to the Linux implementation already being there, I just assumed that the API now basically has to be whatever the Linux implementation conforms to.

[ianlancetaylor]

Yes, there is no need for a proposal to add new system calls for a GOOS that are the same as existing system calls for a different GOOS.