Add Cosmopolitan Libc target #142609

io12 · 2025-06-17T04:55:28Z

This PR adds support for Cosmopolitan Libc, building on prior work from @ahgamut. It depends on support in the libc crate.

Cosmopolitan is a unique target, because some of the system constants (e.g. EINVAL, MAP_ANONYMOUS, O_CLOEXEC) are actually global variables that are set to the correct values at runtime based on the detected OS. This means unfortunately any pattern matching on them like libc::EINVAL => {} must be changed to use if guards instead like e if e == libc::EINVAL => {}. For switch statements in C, the solution was a gcc patch that wasn't upstreamed, so I'm not sure if it makes sense for Rust to upstream this, but I figured I would submit a PR anyway just to make it easier to find.

The code is also a bit messy right now, but it works surprisingly well in its current state.

Testing

Building tests

I built the library tests with

./x test library --target x86_64-unknown-cosmo
./x test library --target aarch64-unknown-cosmo

Then converted the test runners from ELF executables into APE executables with apelink

apelink -o test.com -l $COSMO/.cosmocc/3.9.2/bin/ape-x86_64.elf std-04db952a3d57c5d1.com.dbg

Then copied the APE test runners to different supported platforms and ran them.

Current test results

PASS
- x86_64
  - Linux
  - FreeBSD
  - OpenBSD
  - NetBSD
  - MacOS
- aarch64
  - Linux
  - FreeBSD
FAIL
- x86_64
  - Windows
Untested
- aarch64
  - MacOS
  - Windows

Caution

Concerns (1 active)

libc-system-constants-no-longer-const

Managed by @rustbot—see help for details.

rust-log-analyzer · 2025-06-17T05:18:12Z

The job mingw-check-tidy failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

fmt check
fmt: checked 6077 files
tidy check

thread 'deps (.)' panicked at src/tools/tidy/src/deps.rs:594:24:
cmd.exec() failed with `cargo metadata` exited with an error: error: failed to load source for dependency `libc`

Caused by:
  Unable to update /checkout/library/TODO change to fork

Caused by:
---
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

thread 'main' panicked at src/tools/tidy/src/main.rs:62:49:
called `Result::unwrap()` on an `Err` value: Any { .. }
Command has failed. Rerun with -v to see more details.
Build completed unsuccessfully in 0:00:48
  local time: Tue Jun 17 05:18:01 UTC 2025
  network time: Tue, 17 Jun 2025 05:18:01 GMT
##[error]Process completed with exit code 1.
Post job cleanup.

workingjubilee

Of course, my patch isn’t perfect. It can’t handle some anonymous structs, enums, const ints, or amazing things like using SIGTERM as an array index within an initializer. Rare compiler crashes may still occur in some weird static initializations, or if you try stuffing SIGTERM into a static const int8_t. But I’ve spent a good chunk of time removing obvious counterexamples, and a lot of popular software builds seamlessly. A stringent testing setup will reveal more things to improve.

If that's "seamlessly", then I would hate to know what is "with some difficulty".

I do not think we really want to add a target where none of libc is allowed in const eval, certainly not like this. This is Rust, not C++, and CTFE is a rule when in const {} or any other const context, not a suggestion like it is in constexpr. Many crates do something like this:

const NOT_FOUND: u32 = libc::ENOENT as _;

and then we can't evaluate that, so you're not getting much to begin with. It isn't very common for any one libc symbol, but when considering all the constants that are redefined, it adds up to quite a lot.

And that's assuming that those symbols are even sound, which may be an overly generous assumption.

workingjubilee · 2025-06-17T05:12:04Z

library/std/src/sys/pal/unix/mod.rs

+        x if x == libc::E2BIG => ArgumentListTooLong,
+        x if x == libc::EADDRINUSE => AddrInUse,
+        x if x == libc::EADDRNOTAVAIL => AddrNotAvailable,
+        x if x == libc::EBUSY => ResourceBusy,
+        x if x == libc::ECONNABORTED => ConnectionAborted,
+        x if x == libc::ECONNREFUSED => ConnectionRefused,
+        x if x == libc::ECONNRESET => ConnectionReset,
+        x if x == libc::EDEADLK => Deadlock,
+        x if x == libc::EDQUOT => QuotaExceeded,
+        x if x == libc::EEXIST => AlreadyExists,
+        x if x == libc::EFBIG => FileTooLarge,
+        x if x == libc::EHOSTUNREACH => HostUnreachable,
+        x if x == libc::EINTR => Interrupted,
+        x if x == libc::EINVAL => InvalidInput,
+        x if x == libc::EISDIR => IsADirectory,
+        x if x == libc::ELOOP => FilesystemLoop,
+        x if x == libc::ENOENT => NotFound,
+        x if x == libc::ENOMEM => OutOfMemory,
+        x if x == libc::ENOSPC => StorageFull,
+        x if x == libc::ENOSYS => Unsupported,
+        x if x == libc::EMLINK => TooManyLinks,
+        x if x == libc::ENAMETOOLONG => InvalidFilename,
+        x if x == libc::ENETDOWN => NetworkDown,
+        x if x == libc::ENETUNREACH => NetworkUnreachable,
+        x if x == libc::ENOTCONN => NotConnected,
+        x if x == libc::ENOTDIR => NotADirectory,
        #[cfg(not(target_os = "aix"))]
-        libc::ENOTEMPTY => DirectoryNotEmpty,
-        libc::EPIPE => BrokenPipe,
-        libc::EROFS => ReadOnlyFilesystem,
-        libc::ESPIPE => NotSeekable,
-        libc::ESTALE => StaleNetworkFileHandle,
-        libc::ETIMEDOUT => TimedOut,
-        libc::ETXTBSY => ExecutableFileBusy,
-        libc::EXDEV => CrossesDevices,
-        libc::EINPROGRESS => InProgress,
-        libc::EOPNOTSUPP => Unsupported,
-
-        libc::EACCES | libc::EPERM => PermissionDenied,
+        x if x == libc::ENOTEMPTY => DirectoryNotEmpty,
+        x if x == libc::EPIPE => BrokenPipe,
+        x if x == libc::EROFS => ReadOnlyFilesystem,
+        x if x == libc::ESPIPE => NotSeekable,
+        x if x == libc::ESTALE => StaleNetworkFileHandle,
+        x if x == libc::ETIMEDOUT => TimedOut,
+        x if x == libc::ETXTBSY => ExecutableFileBusy,
+        x if x == libc::EXDEV => CrossesDevices,
+        x if x == libc::EINPROGRESS => InProgress,
+        x if x == libc::EOPNOTSUPP => Unsupported,
+
+        x if x == libc::EACCES || x == libc::EPERM => PermissionDenied,


No, I think not.

The correctness losses here are unacceptable, because changing these to be symbolic, even if it could be done automatically, means that we no longer can benefit from exhaustiveness checking. Thus we cannot do things such as determining if the current match arm is covered by a previous one, or etc., requiring manual rechecking on every single review.

workingjubilee · 2025-06-17T05:12:43Z

library/std/src/os/cosmo/fs.rs

This need not be added for new targets.

workingjubilee · 2025-06-17T05:18:23Z

library/std/src/sys/process/unix/unix.rs

+        let program = self.get_program_cstr();
+        let argv = self.get_argv().as_ptr();
+        #[cfg(target_os = "cosmo")]
+        if libc::IsWindows() && !program.to_bytes().ends_with(b".exe") {


I do not think we want more single-target-specific functions inside the libc crate, either.

bjorn3 · 2025-06-17T10:01:57Z

compiler/rustc_target/src/spec/base/cosmo.rs

+        no_default_libraries: true,
+        max_atomic_width: Some(64),
+        linker_flavor: LinkerFlavor::Gnu(Cc::Yes, Lld::No),
+        stack_probes: StackProbeType::None,


Can't this use inline stack probes?

bjorn3 · 2025-06-17T10:04:38Z

compiler/rustc_target/src/spec/targets/x86_64_unknown_cosmo.rs

+        metadata: TargetMetadata {
+            description: Some("64-bit Cosmopolitan Libc".into()),
+            tier: Some(3),
+            host_tools: Some(true),


Suggested change

host_tools: Some(true),

host_tools: Some(false),

I'm pretty sure rustc can't be compiled for this based on what @workingjubilee already noted.

bjorn3 · 2025-06-17T10:05:22Z

compiler/rustc_target/src/spec/targets/x86_64_unknown_linux_cosmopolitan.rs

bjorn3 · 2025-06-17T10:07:09Z

library/std/src/sys/pal/unix/thread.rs

-            debug_assert_eq!(res, 0);
+            cfg_if::cfg_if! {
+                if #[cfg(target_os = "cosmo")] {
+                    let _ = res;


Why this change?

Urgau · 2025-06-17T16:26:08Z

@rustbot concern libc-system-constants-no-longer-const

I concur with @workingjubilee, loosing the constness of libc system constants is a very big downsides, both in terms of correctness (no more exhaustiveness) but also in terms of usability.

For switch statements in C, the solution was a gcc patch that wasn't upstreamed

If gcc didn't wanted it, then it's not really a solution. I would very much want to see a proper and accepted solution in C before attempting any port in Rust.

tgross35 · 2025-06-18T01:14:20Z

library/std/Cargo.toml

(picking a random file to get an inline comment) new targets should land as no_std first. Once we have that to indicate acceptance in rustc, other ecosystem updates can be considered.

Though maybe, if accepted, this is just easier to support as a no_std target indefinitely? That sidesteps any concerns with libc or hacks in std. It would be possible to publish a cosmo-std crate that drops into the std APIs that are well supported.

I didn't think of this, but it seems like a good idea. In that case, maybe it makes more sense for it to not be part of rustc at all since custom targets can be specified with json files?

The parts you have in compiler/ here can indeed be specified with the target json file, and that is worth trying. I suspect you may eventually need further compiler support to get things working "smoothly", though.

Based on the reactions it seems there are some concerns here. It would probably be a good idea for you to drop in on Zulip and discuss a bit what the best way forward might be https://rust-lang.zulipchat.com/.

io12 · 2025-06-19T02:07:11Z

I agree the changes to make system constants symbolic are probably not worth it. Thanks for the feedback!

This comment has been minimized.

Sign in to view

add cosmopolitan libc support

f21e302

io12 force-pushed the cosmo branch from e2a73de to f21e302 Compare June 17, 2025 05:12

workingjubilee reviewed Jun 17, 2025

View reviewed changes

bjorn3 reviewed Jun 17, 2025

View reviewed changes

compiler/rustc_target/src/spec/targets/x86_64_unknown_linux_cosmopolitan.rs

Copy link

Member

bjorn3 Jun 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Stray file

bjorn3 reviewed Jun 17, 2025

View reviewed changes

rustbot added the S-waiting-on-concerns Status: Awaiting concerns to be addressed by the author label Jun 17, 2025

tgross35 reviewed Jun 18, 2025

View reviewed changes

io12 closed this Jun 19, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add Cosmopolitan Libc target #142609

Add Cosmopolitan Libc target #142609

io12 commented Jun 17, 2025 •

edited by rustbot

Loading

Uh oh!

This comment has been minimized.

rust-log-analyzer commented Jun 17, 2025

Uh oh!

workingjubilee left a comment

Uh oh!

workingjubilee Jun 17, 2025

Uh oh!

workingjubilee Jun 17, 2025

Uh oh!

workingjubilee Jun 17, 2025

Uh oh!

bjorn3 Jun 17, 2025

Uh oh!

bjorn3 Jun 17, 2025

Uh oh!

bjorn3 Jun 17, 2025

Uh oh!

bjorn3 Jun 17, 2025

Uh oh!

Urgau commented Jun 17, 2025

Uh oh!

tgross35 Jun 18, 2025

Uh oh!

io12 Jun 19, 2025

Uh oh!

tgross35 Jun 19, 2025

Uh oh!

io12 commented Jun 19, 2025

Uh oh!

Uh oh!

Add Cosmopolitan Libc target #142609

Add Cosmopolitan Libc target #142609

Conversation

io12 commented Jun 17, 2025 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Testing

Building tests

Current test results

Concerns (1 active)

Uh oh!

This comment has been minimized.

rust-log-analyzer commented Jun 17, 2025

Uh oh!

workingjubilee left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Urgau commented Jun 17, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

io12 commented Jun 19, 2025

Uh oh!

Uh oh!

io12 commented Jun 17, 2025 •

edited by rustbot

Loading