- Where: zoom.us (see Registration below)
- When: September 7, 16:00-17:00 UTC
- Contact:
- Name: Pat Hickey and Bailey Hayes
- Email: [email protected] and [email protected]
If this is your first time attending, please fill out the registration form to receive an invite.
The meeting is open to CG members only. You can join the CG here.
The meeting will be on a zoom.us video conference.
- Opening, welcome and roll call
- Please help add your name to the meeting notes.
- Please help take notes.
- Thanks!
- Announcements
- Submit a PR to add your announcement here
- Proposals and discussions
- Submit a PR to add your announcement here
- Vote: Rebase the Phase Process on CG's process, as presented on 07-27
- Vote: Preview 2 definition, as presented on 07-27
Dan Gohman Conrad Watt Colin Murphy Marcin Kolny Luke Wagner Pat Hickey Petr Penzin Chris Woods Syrus Akbary Jeff Charles Frank Schaffa Ben Green
Pat: A lot of people are at WasmCon today. Since this is a consensus vote, it’s not urgent for people who want to vote yes to be present, because it’s mainly about seeing if anyone wishes to vote no. And since the PRs were posted, we’ve had the opportunity to have a lot of discussions with the people who have raised questions and concerns.
Pat: We don’t believe there will be significant performance regressions on real workloads.
Pat: There are some small overheads today, such as calling malloc in some places where we didn’t before. It’s some small overhead, but we have plans for how we’ll optimize it out.
Pat: Also, I wrote a program that uses WASI to copy files, to read files into memory and write it back out. It’s just one window on WASI performance, but it shows an interesting result. We turned up some interesting low-hanging fruit, but we fixed it, and now it’s close to the same performance as preview1.
Pat: Luke wrote up some great summary comments of the discussion in the PRs.
Chris: We know there will be some issues with the new implementations that we have. It’ll be important to address it, and we added a plan to the PR to measure and address performance within preview3.
Chris: On the Wamr side, there is a workitem to implement a Wit to Witx tool, which is an important piece of the story.
Pat: Wit to Witx tooling is an implementation strategy. It’s not a requirement for Preview 2, we can’t mandate it, but it is worth talking about here. It’s technically possible. We have had a contributor do some Wit to Witx by hand, showing that it can be done, and it’s fairly mechanical. People who chose to go that route will likely wish to automate it with a tool.
Marcin: As of now, the requirement for new WASI proposals is Wit?
Pat: Yes.
Pat: One way to get started on a Wit to Witx translator tool would be to take the existing markdown generator in wit-bindgen and evolve it to emit Witx instead.
Chris: The fact that that tool exists gives us an idea of how much work would be needed. It doesn’t seem like a particularly onerous task.
Marcin: Will wasi-libc need a Canonical ABI interface as well?
Pat: Yes, wit-bindgen has a C header ABI, which is also similar.
Marcin: Perhaps Wamr would want to use that instead of Witx.
Dan: That would work, though that’s for the guest side; generating a C header for the host side would be some additional work, but it’s doable.
Chris: There’s a lot of learning on the Wasmtime side about how to implement the component model; perhaps we should sit down with the Wamr folks to share insights.
Pat: Yes, let’s do that.
Petr: The two features we ran into as we looked into this, threads and component model, don’t seem to have browser buy in. To what extent is this a problem? If someone wants to use WASI on the Web, how could that work?
Pat: That’s a good discussion to have. Rather than dive into that now, let’s table that, and keep talking. We’ll keep talking with browser implementers.
Marcin: My concerns are mitigated, from our perspective. There is a way to stick to our current approach, just updating some APIs. It looks like the performance concerns will be taken into account.
Chris: Likewise, I’m good.
Pat: Let’s do the vote.
Pat: PR549 has passed by unanimous consent.
Pat: PR550 is a follow-on PR. This describes what Preview 2 is.
Pat: Are there any questions or discussions people would like to bring up?
Pat: Seeing none, let’s do the vote.
Pat: PR550 has passed by unanimous consent.
Syrus: For the record, I don’t agree that the component model should be the way forward for WASI, but I don’t want to hold up the vote.
Pat: Ok, we’ll include that in the notes.