Closes#2004.
In porting the bugfix from
6eb852e28f,
I noticed that the LLVM backend was incorrectly generating declaration
initializers (in `callIODeclInitFn`), by assuming the return type of the
initializer is the return type of the declaration. Rather, it must be be
`lean_object`, since the initializer returns an `IO a` value which must be unpacked.
TODO: stop using the `getOrCreateFunction` pattern pervasively.
perform the `create` at the right location, and the `get`
at the correct location.
The goal is to address the regression
```
example : (0 : Nat) + 0 = 0 :=
show 0 + 0 = 0 from rfl
```
introduced by fedf235cba
Note that we should only *try to* unify the types. Otherwise, we would
produce another regression.
```
example : Int :=
show Nat from 0
```
cc @kha
In Lean 4, we have support for typing constraints of the form
```
(?m ...).1 =?= v
```
where the type of `?m ...` is a structure with a single field.
This kind of constraint is reduced to `?m ... =?= ⟨v⟩`
This feature is implemented by the function `isDefEqSingleton`.
As far as I remember, Lean 3 does not implement this feature.
This commit disables this feature if the structure is a class.
The goal is to avoid the generation of counterintuitive instances by
typing inference.
For example, in the example at issue #2011, the following weird
instance was being generated for `Zero (f x)`
```
(@Zero.mk (f x✝) ((@instZero I (fun i => f i) fun i => inst✝¹ i).1 x✝)
```
where `inst✝¹` is the local instance `[∀ i, Zero (f i)]`
Note that this instance is definitinally equal to the expected nicer
instance `inst✝¹ x✝`.
However, the nasty instance trigger nasty unification higher order
constraints later.
Note that a few tests broke because different error messages were
produced. The new error messages seem better. I do not expect this
change to affect Mathlib4 since Lean 3 does not have this feature.
