Make std usage optional for wgpu.

[kpreid]

Connections
Part of implementing #6826. Followup to #7279.

Description
Adds a std feature to wgpu, without which it does not depend on wgpu-core?/std.

Doesn’t build yet, because the WasmNotSendSync trait bound is not met. Further work will be needed, and I’m not sure what yet; I'm publishing this draft PR to help people see what the problem space currently looks like.

Testing
TBD

Squash or Rebase?
TBD

Checklist

• Run cargo fmt.
• Run taplo format.
• Run cargo clippy. If applicable, add:
  • --target wasm32-unknown-unknown
• Run cargo xtask test to run tests.
• If this contains user-facing changes, add a CHANGELOG.md entry.

[bushrat011899]

The test failures look like they're caused by std being disabled when it's required for certain platforms (e.g., Android, VisionOS, etc.). If no_std wgpu isn't supported on those platforms, perhaps add the std feature for those tests in CI?

[bushrat011899]

Actually this just looks like a desync in the wgpu-types requirements for WasmNotSend/Sync and the wgpu cfg option send_sync. Updating the bounds on the various WasmNotSend/Sync traits to match wgpu's send_sync:

// Before
#[cfg(any(
    not(target_arch = "wasm32"),
    all(
        feature = "fragile-send-sync-non-atomic-wasm",
        not(target_feature = "atomics")
    )
))]
pub trait WasmNotSend: Send {}

// After
#[cfg(all(
    feature = "std",
    any(
        not(target_arch = "wasm32"),
        all(
            feature = "fragile-send-sync-non-atomic-wasm",
            not(target_feature = "atomics")
        )
    )
))]
pub trait WasmNotSend: Send {}

Caused the failing CI tasks for pass for me. If that's the fix, might be worth adding a comment indicating the two configuration options need to match.

[kpreid]

@bushrat011899 The problem is, that makes the wgpu-types/std feature non-additive, because it adds a requirement that dependents must satisfy.

[bushrat011899]

Hmm that is a valid point. I assume the inverse, removing std from send_sync in wgpu, isn't possible/sound?

[bushrat011899]

Playing around with this PR, enabling std on the failing platforms is sufficient to pass the --no-default-features tests. Perhaps that is just a mandatory feature for some platforms (could maybe be automatically enabled using target dependencies?). It's not unheard of to require a feature for a crate to compile (see glam with its libm/std feature choice)