Make crate features no-op on incompatible targets

[daxpedda]

This changes crate features to be no-op on incompatible targets instead of not compiling.

• Vulkan is no-op on WASM
• DirectX12 and DirectX11 is no-op on everything but Windows
• Metal is no-op on everything but MacOS and iOS

As far as I'm aware Vulkan is needed on MacOS for Vulkan Portability.
I saw a feature called portable_features that is used for testing purposes. I believe I can actually remove this now?

Depends on #3467.

Fixes #3463 (only WASM).
Fixes #3465.

[daxpedda]

@grovesNL if you could take a look and see if and how you are affected by this.
Any feedback is welcome!

[daxpedda]

What exactly is the point of the emscripten crate feature? Couldn't it be completely replaced by target_os = "emscripten"?
This would also make it not necessary to activate the feature for downstream users.

[grovesNL]

Nice! I thought there was some reason we weren't able to set it up this way originally (maybe @kvark remembers?). Obviously it makes the cfgs slightly messier but it doesn't seem like a big deal since there shouldn't be very many of them anyway.

What exactly is the point of the emscripten crate feature? Couldn't it be completely replaced by target_os = "emscripten"?
This would also make it not necessary to activate the feature for downstream users.

That sounds right - if we don't need a separate feature to conditionally enable it then I think could be replaced by target_os = "emscripten"

[daxpedda]

That sounds right - if we don't need a separate feature to conditionally enable it then I think could be replaced by target_os = "emscripten"

I will do this in a separate PR.

[teoxoy]

Fixes #3463 (only WASM).

Shouldn't it also fix metal?

I saw a feature called portable_features that is used for testing purposes. I believe I can actually remove this now?

I think so.

[teoxoy]

LGTM, thanks for the PR!

We seemed to be moving in this direction already, some cfgs had the target_os (ex windows for dx11/dx12) and also had some dependencies gated by target_os.

This is also more in line with the notion that feature flags should be additive and not cause compilation errors.

[teoxoy]

Could you remove portable_features in this PR as well?

[daxpedda]

Fixes #3463 (only WASM).

Shouldn't it also fix metal?

As far as I am aware it will not, currently metal can't be built on Linux.

Could you remove portable_features in this PR as well?

Will do.

[daxpedda]

I can't remove portable_features right now because I have to put in something instead in the CI, considering this PR I thought it would be appropriate to put in --all-features instead, but that still doesn't work because of emscripten, which should be fixed in #3467.

So either we remove portable_features in a follow-up PR, or I just make this PR depend on #3467.

I tested my assumption in #3488, looks like it works.

[cwfitzgerald]

Let's make the PRs dependent, emscripten should go in soon.

[daxpedda]

Done!

[daxpedda]

This is ready now!

[cwfitzgerald]

This is awesome to finally have!

Does this mean https://github.com/gfx-rs/wgpu/blob/master/wgpu/Cargo.toml#L106-L109 we can enable all features on all platforms?

[daxpedda]

Does this mean https://github.com/gfx-rs/wgpu/blob/master/wgpu/Cargo.toml#L106-L109 we can enable all features on all platforms?

Yes!
And as far as I checked the CI again now this is tested for native.

EDIT: I forgot, it's also tested for web as part of #3467. I remembered wrongly.

[daxpedda]

I was just going through the CI again and I think we might want to test for --all-features a bit more.
Specifically web isn't tested with --all-features and wgpu-hal should maybe be tested separately just to make sure everything works as intended. Should I add this here or should I make a new PR?

[cwfitzgerald]

Let's fix CI in this PR - it's all entangled

[cwfitzgerald]

Some comments, but looking good

[daxpedda]

Just rebased on top of #3505 and #3511.

comments have been addressed