Chris Fallin

Chatted a bit offline with @elliottt about this and we surmised that there may be indirect benefits from this: the changes to CL necessary to fit these constraints plausibly could...

Hmm, yes, right now specifying both is sort of a "user error", which is why the fuzzer doesn't generate that case. I had created the `Stack` constraint basically with GC/stackmaps...

In principle this seems reasonable, but I'll note that this will break a number of links across the Internet -- scattered throughout GitHub comments at least, and e.g. some of...

Yeah I'd be hesitant to introduce a circular dependency in general between repositories. (Perhaps `cranelift-entity` really should be its own little independent library, but that's a slightly bigger question...)

That's a much bigger discussion as well... at the time at least, we had good reasons to not do that: different (lighter-weight) CI here, historical precedent wrt `regalloc.rs`, general modularity...

The switch to 1.0-series versions (bumping major number at each release) don't change anything here, I think; the plan is still for each release to be a semver bump wrt...

@Amanieu: yes, overall, this looks quite reasonable! Thank you very much for working on this. A few thoughts: - I like the clean zero-spill-cost approach. One could imagine an alternative...

@Amanieu I think that placing cold blocks at the end of the function in the linear block order should basically *just work*, as you say. The two issues in the...

@jameysharp would you mind also doing a Sightglass run on this? It *should*, from first principles, mainly result in speedups; but I'm getting some weird results when trying it on...

OK, cool; that's what I had understood, I was mostly just wondering if there was some sort of doc-comment we could add to clarify. I think it's probably fine as-is....