The compiler frontend implemented in C++ is eagerly inlining local
functions. The new test would take an absurd amount of time without
the new hack.
We remove this hack when we re-implement the compiler frontend in Lean.
The scope of the expr2ctor cache updates was incorrect.
This bug affects code of the form
```
let x := C.cases_on y ...; K[x]
```
when we try to float the `cases_on` application, and the continuation
`K[x]` contains another `cases_on` application with major `y`.
The new test exposes the bug.
This commit also fixes the case where the continuation `K[x]` projects `y`.
Fixes#26
@kha This is another instance of a problem we discussed last summer.
I guess there are many more instances like this one that we are not handling.
Recall that we want to preserve types at `csimp` because we eventually
want to allow users to use equational theorems as rewriting rules during
compilation.
However, some of the transformations that `csimp` perform do not
preserve typeability in CIC.
Moreover, some of the optimizations require type inference.
I see the possible long term solutions:
1- Erase types and proofs as soon as possible. The main drawback here is
that we would have to develop an approach for translating Lean theorems
into valid rewriting rules for lambda pure. For example, the following
theorem should not be used as a rewriting rule after we erase types.
```
forall (xs : List Unit) (f : Unit -> Unit), List.map f xs = List.map id xs
```
BTW, I don't want to abandon the idea of using lemmas as rewriting rules in
the compiler.
2- Go over `csimp` and other compiler steps and make sure they work
properly even when `type_checker::infer_type` fails.
@kha I keep finding problems with the float `let` inwards
transformation. It is always a nasty interaction between this
transformation and the `reset/reuse` insertion procedure.
The example I used in the new comment can be modified to a
`casesOn` with more than one branch (e.g., `Option.casesOn`).
Suppose we wrote
```
let o : Option Nat := Array.index a i in
let a := Array.update a i none in
Option.casesOn o
none
(fun n, some (Array.update a i (some (n + 1))))
```
In the example above, the compiler will float
`a := Array.update a i none` inwards.
```
let o : Option Nat := Array.index a i in
Option.casesOn o
none
(fun n,
let a := Array.update a i none in
some (Array.update a i (some (n + 1))))
```
Then, adding reset/reuse:
```
let o : Option Nat := Array.index a i in
Option.casesOn o
none
(fun n,
let o := reset o in
let a := Array.update a i none in
let n := n + 1 in
let o := reuse o (some n)
some (Array.update a i o))
```
Similarly to the example in the new comment, the `reset o` will fail since
the array `a` would still have a reference to `o`.
Remarks:
- Haskell also implements float `let` inwards.
- I am not sure how important the float `let` inward transformation is.
- I can see other nasty interactions after we implement user-defined
simplification rules. For example, I guess many users would find the
following lemma to be a good rewriting rule:
```
(Array.update (Array.update a i v) i w) = (Array.update a i w)
```
However, if we use this lemma in the example above, then `Array.update a i none` will be eliminated,
and `reset o` will fail.
@kha The move `x := val` to `casesOn` branch was producing nasty
problems. I documented the issue, and implemented a simple
and sufficient condition for preventing the problem. The approach is very
similar to the one used at `push_proj_fn` at `llnf.cpp`.
I hope this change will not impact existing benchmarks :)
The new `partial def`s allow us to define `fix` in Lean, but the Lean
implementation is not as efficient as the native one. The native one
in C++ use weak pointers to prevent a closure allocation at every
recursive invocation.
This commit also fixes the `fixCore` helper functions that were broken
after we switched to camelCase.
We have updated the test `fix1.lean` to demonstrate the native
implementation is faster. Here are the numbers on my desktop.
```
./run.sh fix1.lean 24
721420279
Time for 'native fix': 816ms
721420279
Time for 'fix in lean': 1.34s
```
`elim_jp1_fn` was incorrectly expanding join points that were used more
than once. The issue is that the `foreach` combinator "may" skip nodes
that have already been visited.
`whnf_core(e)` uses `whnf` to reduce the major premise of recursors and
projections, and `whnf` unfolds arbitrary definitions.
This commit adds a new option (`cheap`) to `whnf_core`. When
`whnf_core(e, true)` is used, the type checker will not unfolding
definitions when reducing the major premises.
