Nullability attributes need some love

[litherum]

I had to make quite a bit of changes to make the nullability attributes actually compile. I eventually got them to work in https://github.com/WebKit/WebKit/pull/14652/files#diff-1a3f28db146a69a5330a62cca964426d2794d35086c2d2e9f9a23e493a54d433

• The attribute is in the wrong place. Right: char const * WGPU_NULLABLE label; Wrong: WGPU_NULLABLE char const * label;
• You probably want to add a place to stick _Pragma("clang assume_nonnull begin") and _Pragma("clang assume_nonnull end") so you don't have to annotate literally everything
• Even when using assume_nonnull, there are still some places that need an explicit WGPU_NONNULL. I don't quite understand why. The docs say "More complex pointer types ... must be explicitly annotated."
• wgpuAdapterEnumerateFeatures() and wgpuDeviceEnumerateFeatures() need to have their argument be nullable. These functions are designed to be called twice: once to let the caller know how much storage to allocate, and then a second time with the actual store. That first call needs to support nullptr.

On the upside, though, after getting these attributes to compile, they found 2 real bugs in our implementation!

---

Original issue: #119

EDIT(kainino0x): adding onto that list from #182 (review) and others:

• ChainedStruct pointers should be nullable
• Typedef'd pointer types (like callback types) should be annotated
• Objects types (which are opaque pointers) should be annotated
• Return types should be annotated (like wgpuCreateInstance())

[kainino0x]

• You probably want to add a place to stick _Pragma("clang assume_nonnull begin") and _Pragma("clang assume_nonnull end") so you don't have to annotate literally everything
• Even when using assume_nonnull, there are still some places that need an explicit WGPU_NONNULL. I don't quite understand why. The docs say "More complex pointer types ... must be explicitly annotated."

Since we're generating the header anyway I kind of prefer annotating everything. It's noisy but I think it's better than having to go learn the rules about exactly what is and isn't affected by assume_nonnull...

[kainino0x]

Nullability attributes are broken so they can't be used at all right now. It is quite a bit of work to get them fully working so I would like to declare this non-breaking, can be fixed after 1.0.

[kainino0x]

This is basically the same as the WGPU_*_ATTRIBUTE macros which are missing in a few places (at least WGPUStringView) but we've decided don't need to work for 1.0 (#179).

[kainino0x]

Dec 12 meeting:

• Fixing WGPU_NULLABLE, introducing WGPU_NON_NULL, and also fixing some WGPU_*_ATTRIBUTEs. Postpone to after 1.0?
• KN: The attributes in particular will require someone to actually write obj-c bindings to make sure that they're working correctly. But nullability also not very important for c/c++ users.
• OK