T-spec meeting 2024-07-25

--- title: T-spec meeting 2024-07-25 tags: ["T-spec", "meeting", "minutes"] date: 2024-07-25 discussion: https://rust-lang.zulipchat.com/#narrow/stream/399173-t-spec/topic/Meeting.202024-07-25 url: https://hackmd.io/HqqBD0sYRGmdtEBZ0VC5vQ --- Attendees: Joel Marcey, Pierre-Emmanuel Patry, Monadic Cat, Connor Horman, Eric Huss, Urgau, TC, pnkFelix, Sid Askary Agenda: * Inline Assembly Review Closure * ABI Chapter Progress * Chapter List Review * What's next? ## Inline Assembly Review Closure [reference#1523](https://github.com/rust-lang/reference/pull/1523) Joel: How can we get this chapter into a place where it can be merged? TC: We need to optimize for making review easier by separating out changes and additions to content, however necessary they may be, from the reformatting to the greatest extent possible. This is the same standard we'd expect anywhere else in the project. We'd reject outright some huge `rustc` refactoring that was intermixed with behavioral changes and new features, regardless of how significant the purported improvements. TC: As it is, this review is very costly in time. I don't see an alternative to carefully reading the current content, top to bottom, and then Connor's proposed content, and then going point by point. And this is particularly long and involved chapter. Joel: Feels like we are a bit more hamstrung with review since we are dealing with a live, public website (the Reference) vs. a greenfield specification. TC: Disagree. Merging these things under lower standards privately would have just put us in the same situation we're in now, but later, and with more errors piled on. Eric: +1. Joel: Disagree and commit. Probably a non-sequitor anyway since we are where we are. Connor: Big difference is that with the spec repo it was mostly semantic content. Reference is both semantic and formatting. Felix: Do we need to look at the old content at all when reviewing? TC: It's content specific. For inline assembly, yes, for me. For other things, maybe not as much. Joel: Didn't Amanieu, the content expert, already approve this? Eric: Yes, but I am not satisfied with that. TC: Agree with Eric. Amanieu's job is to hold high standards in other places in the project. It's our job to hold high standards here, and we're fortunate he took the time to look at it at all. We shouldn't read more into his review than that he looked it over for egregious errors. Joel: In the interest of time, I think we should give Connor and others freedom to work on other chapters even if previously written chapters haven't been fully-reviewed and merged yet. Felix: Worried about totally committing to minirust at this point in time. Eric: Not fully on board with NDR (No Diagnostic Required). Can't we just call it undefined behavior? > Monadic Cat: The problem with calling IF-NDR "UB" is that the opsem people are trying to make UB be something that's a local behavior in a hypothetical abstract machine, even if that AM can time travel. Felix: We need to figure out how to handle NDR vs. Undefined Behavior. TC: That we're trying to figure it out right now is the process problem. Our plan was to do reformatting only as the first step. That's why we thought we could get all the chapters done in weeks or a small number of months. We're not following our plan, which implies we're not going to get the results we planned in the timeline we planned them. So we need to either get back on plan or accept a new plan and the expected results and timeline of that one. ## ABI Chapter Progress [reference#1545](https://github.com/rust-lang/reference/pull/1545) ## Chapter List Review [Rust Specification Reference Rewrite - Chapter Order](/T5cTU7N6RKKLHr4MIx-lbQ) ## What's next? Joel: Out for two weeks August 8, 15. Eric: Out for 2 weeks August 1, August 8.