*Top-level* Record definitions need accompanying type signatures

[jamesmckinna]

Working on #2111 exposes a previously uncaught error: under Data.Tree.AVL, a number of modules (attempt to) declare 'local' instances of StrictTotalOrder without declaring a suitable type signature for them, which worked perfectly fine as long as this record type was a leaf in the Algebra.Bundles hierarchy...

... but adding (Is)DenseLinearOrder as a sub(structure)/bundle of (Is)StrictTotalOrder has now broken that.

Remedy: identify all record definitions without a type signature, and supply it appropriately.
UPDATED: issue only applies to top-level definitions; within the scope of an enclosing record, the type of the field being defined as a sub-record ensures that a suitable type is being supplied to the typechecker.
UPDATED: I've attempted to identify any other instances in the library, but it seems just to have been these 'historical' examples. Suggest closing the issue now/for the time being.

[MatthewDaggitt]

Ooh interesting. Haven't come across that sort of bug before!

[jamesmckinna]

So the bug expressed itself in terms of unsolved Level meta variables... my diagnosis in terms of leaves in the hierarchy is post hoc inference, and there may be other possibilities, but inserting the type declaration certainly solves/d the instances that came up.

And yes, a weird one! Added label:bug because this is a 'latent' bug, and one to program defensively against.