WIP: Try to extend CI to 9.2.1

[michaelpj]

No description provided.

[michaelpj]

Looks like we're blocked on at least:

• New Hackage release supporting some-1.0.3 obsidiansystems/dependent-sum#66
• Seems not to want to compile on ghc 9.2.1 obsidiansystems/constraints-extras#45

[michaelpj]

I've pulled out the bounds changes to another PR, but this won't work for a bit.

[Ailrun]

@michaelpj For constraint-extras, it looks like their HEAD already supports GHC 9.2, and just the actual release haven't been cut yet. (At least their CI works: https://github.com/obsidiansystems/constraints-extras/runs/4389445166?check_suite_focus=true)

[Bodigrim]

Looking at this and similar tickets for HLS, it seems that dependent-map subtree (dependent-sum, constraints-extras, dependent-sum-template, th-extras) causes constant delays, is it? Could it be replaced with typerep-map? Or maybe roll out a hand-written implementation of a type, isomorphic to DMap SMethod?

[michaelpj]

Yes, I think that's a good suggestion. I'd wondered about doing that, but I don't currently have a handle on how it's used. I'll take a look at what you suggested.

[michaelpj]

Removing dependent-sum and friends isn't trivial. We use GEq in a few places apart from the dependent map in lsp, and while we could just depend on some to get GEq, I really wouldn't want to write the instances by hand (big datatypes!), which pushes us back into the arms of dependent-sum-template.

[michaelpj]

Okay, now I think it's worth having a look at whether we can indeed get rid of dependent-map. It seems that we use it "properly" in a way that can't be obviously replaced by typerep-map. So hand-rolling a simple alternative might be the way to go.

[jneira]

Okay, now I think it's worth having a look at whether we can indeed get rid of dependent-map. It seems that we use it "properly" in a way that can't be obviously replaced by typerep-map. So hand-rolling a simple alternative might be the way to go.

we could leverage changes in lsp? or hls ones are highly unrelated with?

[michaelpj]

I don't think I could get rid of the usage in lsp, or at least I don't understand it enough to say yet. So easier to replace the datastructure at the moment.

[michaelpj]

Done in #384