Stream: wit-bindgen

Topic: status of resource

Nathaniel McCallum (Nov 10 2022 at 14:22):

The wit documentation defines the resourceitem but this appears to be removed from wit-bindgen. What is the status of this?

Dan Gohman (Nov 14 2022 at 18:38):

The status is that resources are a very important feature but are not yet described by the component-model spec. wit-bindgen had a prototype implementation, but it's been removed in order to simplify other component implementation work, and because the semantics of handles are expected to be a little different from what was prototyped.

Dan Gohman (Nov 14 2022 at 18:39):

Luke and others are actively working on the new resource+handle design in the component model, as it's a high priority for a lot of use cases.

Nathaniel McCallum (Nov 16 2022 at 19:35):

@Dan Gohman Is the expectation to install wit-bindgen from git?

Lann Martin (Nov 16 2022 at 19:46):

Yes

Lann Martin (Nov 16 2022 at 19:47):

Intentionally (I think) not on crates.io because of how unstable it is known to be

Nathaniel McCallum (Nov 22 2022 at 18:20):

@Alex Crichton I see that world file syntax has landed. Can you give me a pointer to the documentation for that syntax?

Nathaniel McCallum (Nov 22 2022 at 18:23):

I found this: https://github.com/WebAssembly/component-model/blob/8a5669d4b6e534f3cd70a4b90f482ab94e4e891d/design/mvp/WIT.md#item-world

Nathaniel McCallum (Nov 22 2022 at 18:25):

But that document still has resources, etc in the syntax.

Alex Crichton (Nov 22 2022 at 18:29):

I'd recommend reading tests and such for now, it's not documented yet

Nathaniel McCallum (Nov 22 2022 at 18:36):

Has the use keyword been removed?

Peter Huene (Nov 22 2022 at 18:50):

yes, as part of the implementation of parsing worlds; much like resource, it was removed in anticipation of doing it "for real" with a proper spec

Nathaniel McCallum (Nov 22 2022 at 18:59):

Is it possible to use a type defined in one interface in another interface?

Alex Crichton (Nov 22 2022 at 19:01):

With use being temporarily removed, no, you'll need to copy/paste

Alex Crichton (Nov 22 2022 at 19:01):

It's expected to not take too long before re-adding use

Nathaniel McCallum (Nov 22 2022 at 19:02):

Does use also export the values specified by use?

Alex Crichton (Nov 22 2022 at 19:26):

Sorry I don't know how best to answer that since use isn't currently implemented

Nathaniel McCallum (Nov 22 2022 at 19:52):

@Alex Crichton I'm implementing it now. So I'm asking what the preferred implementation should be. Sorry for lack of clarity.

Alex Crichton (Nov 22 2022 at 19:53):

Have you talked with Luke or Brian or others? There's a number of design questions about use which are unresolved which were why the initial implementation was removed and deferred temporarily.

Nathaniel McCallum (Nov 22 2022 at 19:53):

Nope. But I'm happy to chat.

Nathaniel McCallum (Nov 22 2022 at 19:54):

@Alex Crichton Which Brian?

Alex Crichton (Nov 22 2022 at 19:55):

this one -- https://github.com/fibonacci1729/

Brian (Nov 22 2022 at 19:55):

:wave: Hey Nathaniel, happy to chat

Alex Crichton (Nov 22 2022 at 19:55):

I think we'll try to meet next week once everyone's back to talk about things, ahead of time I do not personally have a complete picture of what use should look like

Nathaniel McCallum (Nov 22 2022 at 19:55):

@Luke Wagner ?

Nathaniel McCallum (Nov 22 2022 at 19:56):

@Luke Wagner Are you available?

Nathaniel McCallum (Nov 22 2022 at 19:58):

@Brian Available now?

Brian (Nov 22 2022 at 19:58):

I'll be available at the top of the hour

Nathaniel McCallum (Nov 22 2022 at 19:59):

@Alex Crichton @Luke Wagner @Brian https://meet.google.com/kjj-rqog-kni

Alex Crichton (Nov 22 2022 at 19:59):

sorry I'm not personally available to meet about this today, but y'all are of course welcome to

Nathaniel McCallum (Nov 22 2022 at 20:00):

No worries. I think I have enough to go on given my reading of the AST.

Luke Wagner (Nov 22 2022 at 20:17):

I'm happy to jump in now or later.

Brian (Nov 22 2022 at 20:36):

Just chatted with Nathaniel and he is working to re-add use to enable sharing types between interfaces defined in the same *.wit file, improving the usability of *.wit today. This makes sense to me since most of the design questions around use arise from wits that are spread across multiple files.

Nathaniel McCallum (Nov 22 2022 at 21:28):

@Brian Having now torn a bunch of this apart, I think defaulting to "private" imports doesn't really make sense. You basically only import types for the purposes of re-exporting them OR using them as params/results for functions. In the later case, the types must be public otherwise the functions aren't callable.

Brian (Nov 22 2022 at 21:47):

@Nathaniel McCallum So instead of what we discussed (i.e. re-exporting via a type alias),
are you suggesting something like:

interface a {
    type x = string
}

interface b {
    use { x } from a
}

interface c {
    use { x } from b
}

Nathaniel McCallum (Nov 22 2022 at 22:14):

@Brian Yes, precisely.

Philpax (Jan 11 2023 at 12:06):

Hi! Apologies for resurrecting this thread, but I was wondering if there had been any progress on same-file type sharing between interfaces. I'd like to share types between my imports and exports, but I'm guessing this isn't supported yet: error: expected type or func, found keyword use` for a use * from common in the interface`.

Will this be implemented soon, and if not, what's the best way to handle type-sharing? (i.e. before updating, I had duplicate types that I had to manually merge together in the host code. Is that still necessary?)

Cheers!

Dan Gohman (Jan 11 2023 at 12:09):

Progress is being made! https://github.com/bytecodealliance/wasm-tools/pull/867

Philpax (Jan 11 2023 at 13:20):

Oh wow, that's awesome, thank you! Looks promising; I'll see if I can postpone this update until that lands.

Last updated: Nov 22 2024 at 16:03 UTC