Remove old generics and generics-rep record support

[paf31]

This removes

• the old generic deriving mechanism
• deriving for records in generics-rep

It needs some changes to generics-rep, and we'd probably want to deprecate generics too.

This is breaking, so it would go into 1.0 or 0.12.0.

Finally, this means we need to figure out how to derive Eq, Show, Ord, etc. for records using RowToList, which means moving some things around in the core libraries.

[hdgarrood]

I was expecting that we wouldn't derive Eq, Show, Ord etc, but rather just define them as normal instances; is that not the case? What are the unresolved questions around instances such as Show (Record a)?

[hdgarrood]

Also, this seems fairly significant, so I think it might be sensible to call the first release containing this v0.12.0, just in case we later notice some problem with it which we want to fix, but fixing it requires another breaking change.

[paf31]

Yes, sorry I meant we should write instances for those.

And yes, I think we do need a 0.12.0 release separate from 1.0.0 at this point.

[hdgarrood]

Is there anything stopping us from writing those instances now, and releasing updated core libraries with those instances, so that people can use them with the 0.11.x compiler?

[paf31]

Yes, we'd need to move RowToList somewhere where the new instances in Prelude could use it.

[paf31]

Fixes #2911.

[hdgarrood]

Would it make sense to add a test which ensures that code which attempts to derive an old-style Generic instance fails with the correct error? I suppose we probably have a test for that already.

[hdgarrood]

Would it make sense to add a test which ensures that code which attempts to derive an old-style Generic instance fails with the correct error? I suppose we probably have a test for that already.

[paf31]

@hdgarrood Done, thanks! I'll merge once CI passes.

[joneshf]

Would it be possible to add a Fail instance to generics/generics-rep describing this particular change and the resolution?

[paf31]

How would it work? The compiler won't derive instances for records any more, so the issue is not that we don't want an instance to exist, but that we want other instances to exist (e.g. Show (Record r)) now.

[joneshf]

I see. What sort of message do you get when you try to derive now?

[paf31]

No instance for YourClass { ... }

[fsoikin]

What's the official way of migrating old Data.Generic-based code to Data.Generic.Rep? Is there, perhaps, a stopgap instance New.Generic a r => Old.Generic a implemented somewhere?

[paf31]

You could try using this code. If it doesn't get merged, I'll probably close the PR and make a generics-compat repo.

[fsoikin]

Got it, thanks @paf31.

One question, out of curiosity though: a comment on that PR states that "For standard data structures, you can simply write derive instance deepFoo :: Deep Foo in the module they are declared, and the instance methods will be filled in for you." How does that work exactly? PureScript can't do DeriveAnyClass, can it?

[paf31]

I may have copied that incorrectly from the original docs 😄

[fsoikin]

Ok, got it.