FD_ISSET: take a *const fd_set instead of *mut fd_set #1725
Conversation
|
r? @JohnTitor (rust_highfive has picked a reviewer for you, use r? to override) |
|
While I was researching my nix PR, I came across this page about the const-ness of that argument. I'm not enough of an OS expert to opine on whether the arguments are convincing, but thought I'd pass along in case it's useful. |
|
Uhm, it's reasonable indeed but I'd prefer as-is since it may cause some confusion by difference with the original definition. |
|
By "original definition" do you mean existing versions of libc, or what's documented in C libraries? AFAIK the only place in C that describes the pointer as mutable is in some man pages. But that's a misnomer since the function is really a macro, and its arguments don't have constedness. And some operating systems document |
|
Fair enough, but on the other hand, it makes the current signature stricter, I think we shouldn't do that without any strong reason or announcement. |
|
Apparently libc does not object to changing the signatures of other functions, even when that causes downstream breakage. Since that's the case, why not merge this PR? |
|
☔ The latest upstream changes (presumably #1975) made this pull request unmergeable. Please resolve the merge conflicts. |
|
Triage: Re-visiting, I still, however, don't see much motivation for this. |
|
The motivation is const-correctness. Right now for example, Nix forces users to use a mutable reference in a place where an immutable reference should suffice, just because libc defines |
FD_ISSET does not modify its fd_set argument, so it may as well take a const pointer. AFAICT the only reason to take a *mut pointer is because the Linux man page documents it that way (though since glibc implements it as a macro, the constedness is undefined). But since libc implements it directly rather than calling a (nonexistent on most platforms) C function, we're defining the API ourselves.
|
A |
|
Thanks for clarifying both! |
|
@bors r+ |
|
📌 Commit 3eafb3b has been approved by |
|
☀️ Test successful - checks-actions, checks-cirrus-freebsd-11, checks-cirrus-freebsd-12, checks-cirrus-freebsd-13 |
There was never any good reason to use mutable receives in the first place. Nix originally did so because libc defined FD_ISSET as taking a mutable receiver, which it did based on an inaccurate Linux man page. But that's fixed now. rust-lang/libc#1725
FD_ISSET does not modify its fd_set argument, so it may as well take a
const pointer. AFAICT the only reason to take a *mut pointer is because
the Linux man page documents it that way (though since glibc implements
it as a macro, the constedness is undefined). But since libc implements
it directly rather than calling a (nonexistent on most platforms) C
function, we're defining the API ourselves.