0.19.0 · wasmtime · Zulip Chat Archive

Stream: wasmtime

Topic: 0.19.0

Peter Huene (Jul 15 2020 at 19:32):

Thoughts on a release soon? We've had reference types go in, an important fix for Windows (crash on some trap unwinds), and other work. The krustlet folks would like to remove a pin to a wasi-common fix commit in their Cargo.toml too.

Chris Fallin (Jul 15 2020 at 19:35):

If it can wait until tomorrow, we have a bugfix for aarch64 / regalloc.rs reftypes that should go in tomorrow once we bump regalloc.rs (SpiderMonkey vendors a pinned git rev so it doesn't matter there, just considering commandline wasmtime)

Chris Fallin (Jul 15 2020 at 19:35):

If not, it's no big deal

Chris Fallin (Jul 15 2020 at 19:36):

... actually we haven't verified all the bits needed for commandline wasmtime to use reftypes on aarch64 anyway, so scratch that. No objections from me :-)

Peter Huene (Jul 15 2020 at 19:39):

i think the krustlet people are just looking for a timeline as they're blocked on publishing a new version until we do, so if there's important stuff on the horizon that should get in, let's do so

Alex Crichton (Jul 15 2020 at 19:39):

I'll write up some release notes!

Dan Gohman (Jul 15 2020 at 19:41):

A release soon would be great

Dan Gohman (Jul 15 2020 at 19:42):

I haven't found time to work on the cargo publish scripts yet; would anyone be interested in helping with that?

Alex Crichton (Jul 15 2020 at 19:56):

I'd also be down to help with that!

Andrew Brown (Jul 15 2020 at 21:08):

@Alex Crichton, the SIMD support on x86 might be worth mentioning, even if it is in the old backend.

Alex Crichton (Jul 15 2020 at 21:09):

certainly!

Alex Crichton (Jul 15 2020 at 21:18):

https://github.com/bytecodealliance/wasmtime/pull/2027

Wasmtime 0.19.0 and Cranelift 0.66.0 by alexcrichton · Pull Request #2027 · bytecodealliance/wasmtime

This commit updates Wasmtime's version to 0.19.0, Cranelift's version to 0.66.0, and updates the release notes as well.

Alex Crichton (Jul 15 2020 at 21:23):

currently I'm thinking we can merge that and publish it tomorrow

Chris Fallin (Jul 15 2020 at 21:37):

Re: above regalloc version bump, it should happen tomorrow morning (PDT) but if it doesn't, don't let me block the release!

Alex Crichton (Jul 15 2020 at 21:38):

@Chris Fallin is it expected that it will enable all reference types tests on aarch64?

Chris Fallin (Jul 15 2020 at 21:40):

Ah, so, interesting, I have not removed the ignore-directive for those reftypes tests. I've relied on the SpiderMonkey-side tests to drive my impl (along with regalloc-side tests and fuzzing).

Chris Fallin (Jul 15 2020 at 21:40):

So, I don't actually know

Chris Fallin (Jul 15 2020 at 21:40):

If the answer is yes (I'm about to find out), maybe it makes sense to release tomorrow? If not, don't let me hold you back now

Chris Fallin (Jul 15 2020 at 21:43):

... The answer is apparently that yes, reftype tests in wasmtime do work on aarch64. The regalloc bump we're waiting for fixes an incorrect stack map in some corner cases, so we should wait for that before claiming reftypes work on aarch64. In the meantime I'll submit a one-line PR to remove the test-ignore directive :-)

Alex Crichton (Jul 15 2020 at 21:45):

wow nice!

Alex Crichton (Jul 15 2020 at 21:45):

I'm surprised the stack walking all "just works", that's pretty impressive

Alex Crichton (Jul 15 2020 at 21:47):

one thing I think we may want to hold off 0.19.0 for is a prototype of reference types support in the other-language-bindings as well

Alex Crichton (Jul 15 2020 at 21:47):

to make sure that we've got all the support needed in the C API

Alex Crichton (Jul 15 2020 at 21:48):

@fitzgen (he/him) looking at the Go stuff, one thing we may be missing is a way to go from wasm_func_t to wasm_val_t?

fitzgen (he/him) (Jul 15 2020 at 21:48):

yeah, I have that in a local commit

Alex Crichton (Jul 15 2020 at 21:48):

ah ok cool

fitzgen (he/him) (Jul 15 2020 at 21:49):

was going to make a PR once the wasmtime-py stuff is ready

Alex Crichton (Jul 15 2020 at 21:49):

makes sense yeah, no worries

Chris Fallin (Jul 15 2020 at 21:59):

Just commented on the PR but: gc smoke-test fails, so cancel the partyhats and allow us new-backend folks to slink back to the laboratory for more work :-)

Chris Fallin (Jul 15 2020 at 21:59):

I'm guessing the build.rs tests just hit the opcodes, not the walking?

Chris Fallin (Jul 15 2020 at 21:59):

In any case, I can debug this more later; don't hold up the release for us

Alex Crichton (Jul 15 2020 at 22:03):

aww

Alex Crichton (Jul 15 2020 at 22:03):

it's true though that gc is not required for correct wasm semantics, just for reasonable embedder semantics

Chris Fallin (Jul 15 2020 at 22:05):

(FWIW, the signature of the smoke-test failure is an assert in regalloc and makes me think it may be fixed by Julian's fuzzbug fix that will land in regalloc tomorrow; so we may be close)