fix: compilation on wasip2

[crepererum]

Before this change, pyo3 cannot be compiled for wasip2:

❯ PYO3_CROSS_PYTHON_VERSION=3.14 PYO3_CROSS_LIB_DIR=... cargo check --target=wasm32-wasip2
   Compiling pyo3-ffi v0.25.1 (/home/mneumann/src/pyo3/pyo3-ffi)
   Compiling pyo3 v0.26.0-dev (/home/mneumann/src/pyo3)
error[E0658]: use of unstable library feature `wasip2`
  --> src/conversions/std/osstr.rs:26:25
   |
26 |             let bytes = std::os::wasi::ffi::OsStrExt::as_bytes(self);
   |                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

error[E0658]: use of unstable library feature `wasip2`
  --> src/conversions/std/osstr.rs:90:17
   |
90 |                 std::os::wasi::ffi::OsStrExt::from_bytes(fs_encoded_bytes.as_bytes(ob.py()));
   |                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

For more information about this error, try `rustc --explain E0658`.

(you can also use a nightly compiler and/or add --features=nightly, this won't change anything)

Now it can be compiled with:

PYO3_CROSS_PYTHON_VERSION=3.14 PYO3_CROSS_LIB_DIR=... cargo +nightly check --target=wasm32-wasip2 --features=nightly

It's a bit sad that this requires nightly, but until rust-lang/rust#130323 is fixed, we cannot do much about it. A similar workaround was chosen by tempfile.

[bschoenmaeckers]

I think we can make this work on stable. As OsStr coming from the wasip2 is guaranteed to be valid utf8. So OsStr::to_str should almost never fail. And if it does fail it is clearly a bug.

[crepererum]

Good point. I think this should work, but can you check that this makes sense to you?

OS => bytes

let bytes = std::os::wasi::ffi::OsStrExt::as_bytes(self);

in WASI should just be

let bytes = self.as_encoded_bytes(); // WASI strings are UTF8

?

bytes => OS

let os_str: &OsStr =
    std::os::wasi::ffi::OsStrExt::from_bytes(fs_encoded_bytes.as_bytes(ob.py()));

should be

let os_str: &OsStr = OsStr::new(
    str::from_utf8(fs_encoded_bytes.as_bytes(ob.py()))?
);

(using this error conversion)

[bschoenmaeckers]

Using self.as_encoded_bytes() sounds somewhat scary. As the docs say that it is an unspecified encoding that only roundtrips well on the same platform/rust version. Maybe this is not a big issue but I think we may want to play safe and error when it is not valid utf-8.

Maybe an other maintainer has some more insight in possible failure cases.

[davidhewitt]

If we expect utf8 to always be the case, perhaps it's good enough to panic in this instance for now? I.e. to OS => bytes conversion can be

let bytes = self.to_str().expect("wasi strings are UTF8").as_bytes();

[crepererum]

This now uses the proposed stable-Rust conversation routines and is ready for another round of reviews.

[davidhewitt]

Thanks for reporting and fixing!