The unifier used to implement the `cases` tactic should not discard
equations of the form `x = t` and `t = x` using proof irrelanvance.
The new test demonstrates the issue. The unifier was reaching the
state
```
x : Conw Con.nil
|- x = Conw.nilw -> x = Conw.nilw
```
and discarding the equality instead of substituting `x`
because `x` and `Conw.nilw` are definionally equal due to
proof irrelevance.
@javra Do you have more complicated examples that were being
affected by this issue?
The new class specifies an interface for saving and restoring the
backtrackable part of the state.
This commit also fixes a few issues.
- `commitWhen` at `LevelDefEq` was defining a checkpoint for
the `isDefEq` methods, and it affects how postponed universe
constraints are handled. However, the name suggests it is
similar to `commitWhenSome?`, and consequently it was used
in other places that had nothing to do with `isDefEq`.
So, I renamed it, and provided the generic `commitWhen` at the new
`MonadBacktrack.lean` file.
- We were restoring more state then needed in a few places.
For example, we were discarding all caches.
- At `SyntheticMVars.lean`, we were using the `Meta.commitWhenSome?`
method which does not restore the `Term.State`.
@Kha The default value (false) for `pp.inaccessibleNames == false` help when
visualizing error messages (see test
`hidingInaccessibleNames.lean`). We added this feature after to hide
intermediate variables created by `match_syntax`.
However, this default value confused me in tactic mode. For example,
it will hide a hypotheses `x : Fin 0` if nobody depends on it, but as
a user we want to know we have it since we can close the goal using
it. Thus, I added `withPPInaccessibleNames act`, it executes `act`
using `pp.inaccessibleNames true` if the user did not explicitly set
it. I use this combinator at `FileWorker` and when producing the
`unsolved goals` error message. In all other scenarios, I believe
hiding these inaccessible variables is a good thing.
