Fix nix for Dragonfly #684

mneumann · 2017-07-17T20:56:40Z

No description provided.

Function __dfly_error() is gone from Rust since a long time. Rust's libstd uses the thread_local attribute for errno, but thread_local cannot be used from stable Rust. For this reason I created errno_dragonfly which is a C FFI wrapper for errno.

asomers · 2017-07-17T21:05:59Z

Cargo.toml

@@ -27,6 +27,9 @@ bitflags = "0.9"
 cfg-if = "0.1.0"
 void = "1.0.2"

+[target.'cfg(target_os = "dragonfly")'.dependencies]
+errno-dragonfly = "0.1"


We shouldn't need an extra crate just for this. The basic symbols should all be defined in libc, and nix's won src/errno.rs defines errno_location. Can you merge whatever you need into nix and libc so we don't depend on the extra crate?

No, there is no other possibility!!! I talked with @alexcrichton to make the thread_local attribute attribute stable, and he actually suggested me to write a C extension like errno-dragonfly. libc does not expose the errno variable directly AFAIK.

The problem is that in DragonFly, the errno variable is not a global, but a thread locale variable, so a simple "extern __errno" as used by the other OSes simply does not work. In libstd, errno can be accessed by using "#[thread_local]", as unstable features are allowed for libstd.

If you want to support DragonFly there really isn't any other option to go via C atm.

That's ... sad. I took a stab at this myself. It doesn't look as though we can use #[thread_local] unless we enable it for all builds, not just Dragonfly's. Still, what would you say to moving the C extension into Nix instead of making it its own crate? Cargo just seems like an awfully heavyweight way to manage such a tiny amount of code.

I don't have a problem with this crate staying external, though I'd much rather rely on a crate that had proper semver (i.e. was 1.0+).

asomers · 2017-07-17T21:08:28Z

src/sys/socket/ffi.rs

@@ -20,7 +20,10 @@ pub type type_of_cmsg_len = socklen_t;
 #[cfg(target_os = "macos")]
 pub type type_of_cmsg_data = c_uint;

-#[cfg(not(target_os = "macos"))]
+#[cfg(target_os = "dragonfly")]


I think this should be #[cfg(any(target_os = "dragonfly", target_os = "macos"))]

No, this is correct! type_of_cmsg_data is already defined above as c_uint. Here it is defined for DragonFly as c_int. Below it is defined for all other OSes.

Oh, you're right. I didn't notice the "u".

asomers · 2017-07-17T21:11:12Z

src/sys/wait.rs

@@ -145,27 +145,34 @@ mod status {
          target_os = "netbsd"))]
 mod status {
    use sys::signal::Signal;
+    use libc::c_int;


Why is it necessary to change all of these types?

I don't understand why you use i32 here instead of c_int. The latter is the correct type for those APIs, as waitpid returns a c_int, not a i32. On most platforms these types are identical, but c_int should be the correct one.

Sure, if you want to be pedantically correct. But it's got nothing to do with DragonFly. It you're going to change those types, please do it in a separate commit, not comingled with DragonFly specific changes.

I agree with @asomers here, please remove this changes from this PR.

Susurrus · 2017-07-17T21:25:15Z

There's big changes coming to the FFI declarations in the socket module in #647. I suggest you test that PR first and provide edits there and keep this PR rebased on top of it so it can be merged afterwards.

mneumann · 2017-07-18T22:35:13Z

Ok, I will wait for #647 to land, then rebase on top of that.

asomers · 2017-07-19T04:38:54Z

src/sys/wait.rs

+    }
+
+    #[cfg(not(target_os = "dragonfly"))]
+    pub fn signaled(status: c_int) -> bool {
        wstatus(status) != WSTOPPED && wstatus(status) != 0 && status != 0x13


What does the magic number 0x13 mean? This sounds like a constant that should be in libc.

asomers · 2017-07-19T04:42:17Z

src/sys/wait.rs

@@ -145,27 +145,34 @@ mod status {
          target_os = "netbsd"))]
 mod status {
    use sys::signal::Signal;
+    use libc::c_int;


Sure, if you want to be pedantically correct. But it's got nothing to do with DragonFly. It you're going to change those types, please do it in a separate commit, not comingled with DragonFly specific changes.

Susurrus · 2017-08-01T21:38:12Z

@mneumann Do you want to rebase this and address @asomers review now that #647 landed?

Susurrus · 2017-08-28T00:14:12Z

@mneumann I'd like to go ahead and get these changes except the errno-dragonfly parts merged. We can open that as an issue and defer on resolving it for now. Do you have time coming up here to address this?

Susurrus · 2017-11-17T05:43:26Z

@mneumann I think I'd be fine with errno-dragonfly being a dependency of nix, as I'd like to get this merged. Please remove some of the unnecessary int -> i32 changes you have and rebase and let's get this merged!

mneumann · 2017-11-17T12:17:12Z

Closing in favor of #799

799: Fix nix on Dragonfly r=Susurrus a=mneumann This commit replaces pull request #684. It fixes building `nix` on DragonFly. All tests pass. This requires most recent libc to build: rust-lang/libc#851.

mneumann added 4 commits July 17, 2017 22:37

Fix errno on DragonFly

d3b0d08

Function __dfly_error() is gone from Rust since a long time. Rust's libstd uses the thread_local attribute for errno, but thread_local cannot be used from stable Rust. For this reason I created errno_dragonfly which is a C FFI wrapper for errno.

Fix socket constants for DragonFly

3da33a9

Use c_int instead of i32

afd178a

Do not run aio test cases on DragonFly

d1e19d5

asomers reviewed Jul 17, 2017

View reviewed changes

asomers requested changes Jul 19, 2017

View reviewed changes

Susurrus mentioned this pull request Nov 17, 2017

Migrate all FFI declarations to libc #748

Closed

8 tasks

mneumann mentioned this pull request Nov 17, 2017

Fix nix on Dragonfly #799

Merged

mneumann closed this Nov 17, 2017

bors bot added a commit that referenced this pull request Dec 19, 2017

Merge #799

b68db41

799: Fix nix on Dragonfly r=Susurrus a=mneumann This commit replaces pull request #684. It fixes building `nix` on DragonFly. All tests pass. This requires most recent libc to build: rust-lang/libc#851.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix nix for Dragonfly #684

Fix nix for Dragonfly #684

mneumann commented Jul 17, 2017

asomers Jul 17, 2017

mneumann Jul 18, 2017

asomers Jul 19, 2017

Susurrus Nov 17, 2017

asomers Jul 17, 2017

mneumann Jul 18, 2017

asomers Jul 19, 2017

asomers Jul 17, 2017

mneumann Jul 18, 2017

asomers Jul 19, 2017

Susurrus Nov 17, 2017

Susurrus commented Jul 17, 2017

mneumann commented Jul 18, 2017

asomers Jul 19, 2017

asomers Jul 19, 2017

Susurrus commented Aug 1, 2017

Susurrus commented Aug 28, 2017

Susurrus commented Nov 17, 2017

mneumann commented Nov 17, 2017

Fix nix for Dragonfly #684

Fix nix for Dragonfly #684

Conversation

mneumann commented Jul 17, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Susurrus commented Jul 17, 2017

mneumann commented Jul 18, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Susurrus commented Aug 1, 2017

Susurrus commented Aug 28, 2017

Susurrus commented Nov 17, 2017

mneumann commented Nov 17, 2017