@Kha I was tired of writing `arbitrary _` :)
There 0 places in the stdlib where the type needs to be provided.
If in the future we need to specify the type we can use
`arbitrary (α := <type>)`
@Kha I was writing the following example
```
...
induction as
| nil => cases h -- `h` is an empty type
| cons b bs ih => cases h
| head a bs => exact ⟨[], ⟨bs, rfl⟩⟩
| tail a b bs h1 => ...
```
The current `syntax` assumes the `| cons b bs ih => ...` alternative
is part of the first `cases h`. Forcing the `|` to occur in a column
>= of the corresponding `cases` is not a nice solution since it would
preven us from writing the second `cases` as I wrote it above.
That is we would have to write
```
induction as
| nil => cases h -- `h` is an empty type
| cons b bs ih =>
cases h
| head a bs => exact ⟨[], ⟨bs, rfl⟩⟩
| tail a b bs h1 => ...
```
or
```
induction as
| nil => cases h -- `h` is an empty type
| cons b bs ih => cases h
| head a bs => exact ⟨[], ⟨bs, rfl⟩⟩
| tail a b bs h1 => ...
```
I think the best solution is to use `with` when we have explicit
alternatives with `|`. The new syntax is similar to `match ... with | ...`
That is, we would write
```
...
induction as with
| nil => cases h
| cons b bs ih => cases h with
| head a bs => exact ⟨[], ⟨bs, rfl⟩⟩
| tail a b bs h1 => ...
```
@Kha I implemented the syntax for structure/class that we discussed this morning.
It is much cleaner. See new tests at `struct2.lean`.
I updated the documentation to use it.
The method `getRef` returns the whole term being expanded.
@Kha The method was originially added for reporting error messages.
So, the name is counterintuitive.
