Update Component and WIT import/export syntax

[alexcrichton]

This PR is an implementation of WebAssembly/component-model#193 organized as a series of commits to "sets of crates" within this repository. Each modified crate's tests pass with each successive commit, but the entire repo's tests and fuzzers only pass on the final commit. Like with many WIT changes prior to this one I will preemptively apologize for the size and scale of this commit. Review-wise it's probably best to stick to the core crate of wasmparser and otherwise mostly review tests.

A sumary of changes made in this PR are:

• The component model binary format is updated, and its major version is bumped. "0xc is dead long live 0xd". This is due to the new binary form of imports and exports and the removal of URLs.
• URLs are removed entirely from the binary format and WIT. They're now supplanted with "IDs" which look like foo:bar/baz. In the text format of a component this is (interface "foo:bar/baz").
• WIT syntax has seen a significant number of updates:
  • The concept of a "document" is gone. A WIT package is now always representable with a single file.
  • WIT supports top-level use to name remote interfaces locally or rename local interfaces to a different name in the local scope. This use is purely a package organizational tool.
  • The concept of default is now gone. All interfaces and worlds are referred to by "id" which is inferrable from a package's name.
  • All packages must have a package ... directive indicating the name of the package.
  • The syntax of use has been replaced. Now use foo.{...} refers to the name foo in scope, and use foo/bar.{...} is additionally a means of importing from foreign packages.
  • The syntax of imports/exports in worlds now reflects the component model. Kebab-names are not required for interface imports/exports (and no longer allowed). Anonymous interfaces still use the prior kebab-name syntax.

This list of changes isn't necessarily large but the impact of the change is quite large as can be seen from the scale of this commit. This touched quite a lot of fundamental structures and assumptions about how packages/components look. I've done my best to be judicious about all the little design decisions along the way and I'm hopeful that it's all in a pretty good spot to move forward. In terms of correctness I'm feeling pretty confident about this after updating fuzz-test-case-generation and generating a slew of each-more-interesting-than-the-last bugs.

In terms of logistics of landing this, I would like to work on getting Wasmtime and wit-bindgen updated before landing this here to account for any new features/bugs/etc that need ironing out. My hope is to land everything sort of all-at-once to try to keep everything in sync and avoid this repo getting out of date relative to others.

[peterhuene]

This looks fantastic and I'm really happy with how this is shaping up.

Just a few minor issues and nits.

[peterhuene]

With my initial review now completed, I've been thinking more about what it means to be able to specify external dependency versions in WIT directly and how that would integrate with registry-aware tooling.

What's supported in cargo-component today is Cargo.toml can have a table like:

[package.metadata.component.target.dependencies]
# Resolve external dependency with name `a` to version 1.0 of WIT package `foo:bar`
a = "foo:bar@1.0"

and in the target WIT file:

world foo {
   import foo: a.foo
}

and that worked because we assumed a was an external dependency with the current syntax and it was up to the tooling to resolve that symbolic name to a package and version.

Now that we can encode the dependency information directly in the WIT, eliminating the target.dependencies table entirely from Cargo.toml 👍

Regarding versions in external dependency ids, I expect that cargo-component will error on a missing version, as an implicit * feels like the wrong thing to do.

It does make me wonder if we should require a version number for external dependencies when parsing the WIT, though. Given that deps is a workaround in lieu of a functioning registry, it doesn't seem like much of a burden, at first glance, to require that you refer to those deps with a version.

Edit: see next review comment.

[alexcrichton]

Thanks for taking the time to review this! I know it's a lot ...

[peterhuene]

Just the one unresolved comment remaining, otherwise an enthusiastic 👍 from me (I'll be updating cargo-component with this asap).

[alexcrichton]

Thanks again for the review! When WebAssembly/component-model#193 gets closer to merging I'll start the process for merging this and propagating it outwards.

[alexcrichton]

Ok one feature added with the latest commit is that now, thoeretically, world bindgen is now possible without actually defining your own world. For example wit-bindgen --world wasi:http/proxy should now work. This should help use of worlds define in foreign packages.

[alexcrichton]

Ok given what appears to be high-level consensus on WebAssembly/component-model#198 I'm going to merge this and get the ball rolling on releaseing this. There are still some changes being discussed but they're relatively minor and should be easy to roll out as incremental changes ideally.