summaryrefslogtreecommitdiff
path: root/source/_posts/隐藏一些OCaml-Effect的机制,让其语法在精神上更接近delimcc.md
diff options
context:
space:
mode:
Diffstat (limited to 'source/_posts/隐藏一些OCaml-Effect的机制,让其语法在精神上更接近delimcc.md')
-rw-r--r--source/_posts/隐藏一些OCaml-Effect的机制,让其语法在精神上更接近delimcc.md95
1 files changed, 95 insertions, 0 deletions
diff --git a/source/_posts/隐藏一些OCaml-Effect的机制,让其语法在精神上更接近delimcc.md b/source/_posts/隐藏一些OCaml-Effect的机制,让其语法在精神上更接近delimcc.md
new file mode 100644
index 00000000..e6caefb7
--- /dev/null
+++ b/source/_posts/隐藏一些OCaml-Effect的机制,让其语法在精神上更接近delimcc.md
@@ -0,0 +1,95 @@
+---
+title: 隐藏一些OCaml Effect的机制,让其语法在精神上更接近delimcc
+date: 2023-06-28 21:47:16
+tags: [Technique]
+---
+
+
+[delimcc_of_fxhandler这个库](https://github.com/kayceesrk/delimcc_of_fxhandler)在OCaml5的effect handlers上实现了一些delimcc原语(shift/reset, control/prompt这些):
+```ocaml
+ let p = new_prompt () in
+ assert ([] = push_prompt p (fun () ->
+ 1::2::take_subcont p (fun _k -> [])));
+ assert ([1;2] = push_prompt p (fun () ->
+ 1::2::take_subcont p (fun k -> push_subcont k [])));
+ assert (135 =
+ let p1 = new_prompt () in
+ let p2 = new_prompt () in
+ let p3 = new_prompt () in
+ let pushtwice sk =
+ sk (fun () ->
+ sk (fun () ->
+ shift0 p2 (fun sk2 -> sk2 (fun () ->
+ sk2 (fun () -> 3))) ()))
+ in
+ push_prompt p1 (fun () ->
+ push_prompt p2 (fun () ->
+ push_prompt p3 (fun () -> shift0 p1 pushtwice ()) + 10) + 1) + 100);
+
+ print_endline "Success!"
+```
+
+另外, [avsm这里](https://github.com/avsm/ocaml/commits/effect-syntax)可以看到一些OCaml的Effect Syntax进展。
+
+还有 [multi-shot continuations in OCaml](https://github.com/dhil/ocaml-multicont),在这个仓库里面还讨论了一些有趣的问题,例如,OCaml 编译器和runtime会做出一些假设从而进行一些优化,这些优化在使用multi-shot continutation时是不可取的(或完全错误的)。编译器优化导致错误的一个例子是堆到栈的转换,例如:
+```ocaml
+(* An illustration of how the heap to stack optimisation is broken.
+ * This example is adapted from de Vilhena and Pottier (2021).
+ * file: heap2stack.ml
+ * compile: ocamlopt -I $(opam var lib)/multicont multicont.cmxa heap2stack.ml
+ * run: ./a.out *)
+
+(* We first require a little bit of setup. The following declares an
+ operation `Twice' which we use to implement multiple returns. *)
+type _ Effect.t += Twice : unit Effect.t
+
+(* The handler `htwice' interprets `Twice' by simply invoking its
+ continuation twice. *)
+let htwice : (unit, unit) Effect.Deep.handler
+ = { retc = (fun x -> x)
+ ; exnc = (fun e -> raise e)
+ ; effc = (fun (type a) (eff : a Effect.t) ->
+ let open Effect.Deep in
+ match eff with
+ | Twice -> Some (fun (k : (a, _) continuation) ->
+ continue (Multicont.Deep.clone_continuation k) ();
+ continue k ())
+ | _ -> None) }
+
+(* Now for the interesting stuff. In the code below, the compiler will
+ perform an escape analysis on the reference `i' and deduce that it
+ does not escape the local scope, because it is unaware of the
+ semantics of `perform Twice', hence the optimiser will transform
+ `i' into an immediate on the stack to save a heap allocation. As a
+ consequence, the assertion `(!i = 1)' will succeed twice, whereas
+ it should fail after the second return of `perform Twice'. *)
+let heap2stack () =
+ Effect.Deep.match_with
+ (fun () ->
+ let i = ref 0 in
+ Effect.perform Twice;
+ i := !i + 1;
+ Printf.printf "i = %d\n%!" !i;
+ assert (!i = 1))
+ () htwice
+
+(* The following does not trigger an assertion failure. *)
+let _ = heap2stack ()
+
+(* To fix this issue, we can wrap reference allocations in an instance
+ of `Sys.opaque_identity'. However, this is not really a viable fix
+ in general, as we may not have access to the client code that
+ allocates the reference! *)
+let heap2stack' () =
+ Effect.Deep.match_with
+ (fun () ->
+ let i = Sys.opaque_identity (ref 0) in
+ Effect.perform Twice;
+ i := !i + 1;
+ Printf.printf "i = %d\n%!" !i;
+ assert (!i = 1))
+ () htwice
+
+(* The following triggers an assertion failure. *)
+let _ = heap2stack' ()
+```