Stream: git-wasmtime

Topic: wasmtime / PR #9381 Configure WIT feature gates at runtim...

view this post on Zulip Wasmtime GitHub notifications bot (Oct 06 2024 at 10:40):

badeend opened PR #9381 from badeend:unstable-features to bytecodealliance:main:

For extra context, see: https://github.com/bytecodealliance/wasmtime/pull/9276

This PR adds the ability to enable/disable WIT APIs gated by @unstable(feature = ...) at link-time.

Previously, unstable features could only be controlled at compile-time through the features option of wasmtime::component::bindgen!. That option has been removed. Bindings for unstable APIs are now generated unconditionally. Instead, the generated add_to_linker*** functions take an additional LinkOptions parameter if the world/interface has any unstable features. The availability of unstable features is now controlled through that.

The component bindgen generates a fresh LinkOptions type for each interface that needs it and a single LinkOptions type for the entire world, which is a union of all the interface's types. An example is included in the component-macro tests: unstable-features.wit produces:
https://github.com/badeend/wasmtime/compare/245dcab34c2fcae69e0f6c3f2daa15837ea6ff61..4fcfe6f866f36ed663f24199e1744bed3cf0bc2c#diff-ad635965474c257cbafce510463aaf52c1ac813f22168045ae22546d3b3ebfc6

A happy coincidence is that WASI-0.2.1 already includes a tiny but practical first use case: exit-with-code.
I've implemented that function and is gated behind the -Scli-exit-with-code CLI flag.

@sunfishcode, if I interpret https://github.com/WebAssembly/wasi-cli/issues/11 correctly, you were OK with removing the Windows-specific restriction, right? See: https://github.com/badeend/wasmtime/commit/2f48c77867d0bd02c4c910510f491a71da22a0f0

view this post on Zulip Wasmtime GitHub notifications bot (Oct 06 2024 at 10:40):

badeend requested wasmtime-core-reviewers for a review on PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 06 2024 at 10:40):

badeend requested fitzgen for a review on PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 06 2024 at 10:42):

badeend updated PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 06 2024 at 11:00):

badeend edited PR #9381:

For extra context, see: https://github.com/bytecodealliance/wasmtime/pull/9276

This PR adds the ability to enable/disable WIT APIs gated by @unstable(feature = ...) at link-time.

Previously, unstable features could only be controlled at compile-time through the features option of wasmtime::component::bindgen!. That option has been removed. Bindings for unstable APIs are now generated unconditionally. Instead, the generated add_to_linker*** functions take an additional LinkOptions parameter if the world/interface has any unstable features. The availability of unstable features is now controlled through that.

The component bindgen generates a fresh LinkOptions type for each interface that needs it and a single LinkOptions type for the entire world, which is a union of all the options contained in that world. An example is included in the component-macro tests: unstable-features.wit produces:
https://github.com/badeend/wasmtime/compare/245dcab34c2fcae69e0f6c3f2daa15837ea6ff61..4fcfe6f866f36ed663f24199e1744bed3cf0bc2c#diff-ad635965474c257cbafce510463aaf52c1ac813f22168045ae22546d3b3ebfc6

A happy coincidence is that WASI-0.2.1 already includes a tiny but practical first use case: exit-with-code.
I've implemented that function and is gated behind the -Scli-exit-with-code CLI flag.

@sunfishcode, if I interpret https://github.com/WebAssembly/wasi-cli/issues/11 correctly, you were OK with removing the Windows-specific restriction, right? See: https://github.com/badeend/wasmtime/commit/2f48c77867d0bd02c4c910510f491a71da22a0f0

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 14:35):

alexcrichton submitted PR review:

This looks great to me, thanks for this!

In retrospect though given the large number of changes related to add_to_linker what do you think about:

Or perhaps this would only affect the helpers we have in wasmtime-wasi and wasmtime-wasi-http? Basically it looks like there should be a convenience function for "default options" rather than requiring all callers to call it.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 14:35):

alexcrichton created PR review comment:

Could this use a BTreeSet or IndexSet to ensure it always generates a deterministic order of options?

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 17:59):

badeend updated PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 17:59):

badeend submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 17:59):

badeend created PR review comment:

Definitely

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 18:07):

badeend commented on PR #9381:

Don't know what top-level function you're referring to. I think the Interface/world-level add_to_linker functions are the only ones generated by the bindgen?

I have just updated the wasmtime-wasi's functions back to their old signature and added an extra set of _with_options variants. I didn't touch the add_to_linker functions generated by bindgen.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 18:48):

alexcrichton submitted PR review:

Ah sorry for my rambling, this is exactly what I was thinking :+1:

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 19:15):

badeend updated PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 19:15):

badeend commented on PR #9381:

Try again please

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 19:17):

alexcrichton has enabled auto merge for PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 20:17):

badeend updated PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 20:18):

badeend commented on PR #9381:

Sigh.

Try again, again?

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 20:21):

alexcrichton has enabled auto merge for PR #9381.

view this post on Zulip Wasmtime GitHub notifications bot (Oct 07 2024 at 20:43):

alexcrichton merged PR #9381.

Last updated: Nov 22 2024 at 16:03 UTC