catala/compiler/shared_ast/typing.ml

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

817 lines
30 KiB
OCaml
Raw Normal View History

(* This file is part of the Catala compiler, a specification language for tax
and social benefits computation rules. Copyright (C) 2020 Inria, contributor:
Denis Merigoux <denis.merigoux@inria.fr>
Licensed under the Apache License, Version 2.0 (the "License"); you may not
use this file except in compliance with the License. You may obtain a copy of
the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
License for the specific language governing permissions and limitations under
the License. *)
(** Typing for the default calculus. Because of the error terms, we perform type
inference using the classical W algorithm with union-find unification. *)
2022-11-21 12:46:17 +03:00
open Catala_utils
module A = Definitions
module Any =
2022-11-21 12:46:17 +03:00
Uid.Make
(struct
type info = unit
let to_string _ = "any"
2022-11-21 12:46:17 +03:00
let format fmt () = Format.fprintf fmt "any"
let equal _ _ = true
let compare _ _ = 0
end)
()
type unionfind_typ = naked_typ Marked.pos UnionFind.elem
(** We do not reuse {!type: Shared_ast.typ} because we have to include a new
[TAny] variant. Indeed, error terms can have any type and this has to be
captured by the type sytem. *)
and naked_typ =
| TLit of A.typ_lit
2023-01-11 13:23:08 +03:00
| TArrow of unionfind_typ list * unionfind_typ
| TTuple of unionfind_typ list
| TStruct of A.StructName.t
| TEnum of A.EnumName.t
| TOption of unionfind_typ
| TArray of unionfind_typ
| TAny of Any.t
let rec typ_to_ast ?(unsafe = false) (ty : unionfind_typ) : A.typ =
let typ_to_ast = typ_to_ast ~unsafe in
let ty, pos = UnionFind.get (UnionFind.find ty) in
match ty with
| TLit l -> A.TLit l, pos
| TTuple ts -> A.TTuple (List.map typ_to_ast ts), pos
| TStruct s -> A.TStruct s, pos
| TEnum e -> A.TEnum e, pos
| TOption t -> A.TOption (typ_to_ast t), pos
2023-01-11 13:23:08 +03:00
| TArrow (t1, t2) -> A.TArrow (List.map typ_to_ast t1, typ_to_ast t2), pos
| TArray t1 -> A.TArray (typ_to_ast t1), pos
| TAny _ ->
if unsafe then A.TAny, pos
else
(* No polymorphism in Catala: type inference should return full types
without wildcards, and this function is used to recover the types after
typing. *)
Errors.raise_spanned_error pos
"Internal error: typing at this point could not be resolved"
(* Checks that there are no type variables remaining *)
let rec all_resolved ty =
match Marked.unmark (UnionFind.get (UnionFind.find ty)) with
| TAny _ -> false
| TLit _ | TStruct _ | TEnum _ -> true
| TOption t1 | TArray t1 -> all_resolved t1
2023-01-11 13:23:08 +03:00
| TArrow (t1, t2) -> List.for_all all_resolved t1 && all_resolved t2
| TTuple ts -> List.for_all all_resolved ts
let rec ast_to_typ (ty : A.typ) : unionfind_typ =
let ty' =
match Marked.unmark ty with
| A.TLit l -> TLit l
2023-01-11 13:23:08 +03:00
| A.TArrow (t1, t2) -> TArrow (List.map ast_to_typ t1, ast_to_typ t2)
| A.TTuple ts -> TTuple (List.map ast_to_typ ts)
| A.TStruct s -> TStruct s
| A.TEnum e -> TEnum e
| A.TOption t -> TOption (ast_to_typ t)
| A.TArray t -> TArray (ast_to_typ t)
| A.TAny -> TAny (Any.fresh ())
in
UnionFind.make (Marked.same_mark_as ty' ty)
2020-12-14 20:09:38 +03:00
(** {1 Types and unification} *)
let typ_needs_parens (t : unionfind_typ) : bool =
2020-12-30 00:26:10 +03:00
let t = UnionFind.get (UnionFind.find t) in
match Marked.unmark t with TArrow _ | TArray _ -> true | _ -> false
2021-01-14 02:17:24 +03:00
let rec format_typ
(ctx : A.decl_ctx)
2021-01-14 02:17:24 +03:00
(fmt : Format.formatter)
(naked_typ : unionfind_typ) : unit =
2021-01-14 02:17:24 +03:00
let format_typ = format_typ ctx in
let format_typ_with_parens (fmt : Format.formatter) (t : unionfind_typ) =
2020-12-30 00:26:10 +03:00
if typ_needs_parens t then Format.fprintf fmt "(%a)" format_typ t
else Format.fprintf fmt "%a" format_typ t
in
let naked_typ = UnionFind.get (UnionFind.find naked_typ) in
match Marked.unmark naked_typ with
| TLit l -> Format.fprintf fmt "%a" Print.tlit l
| TTuple ts ->
2023-01-08 00:00:57 +03:00
Format.fprintf fmt "@[<hov 2>(%a)@]"
2021-01-14 02:17:24 +03:00
(Format.pp_print_list
~pp_sep:(fun fmt () -> Format.fprintf fmt "@ *@ ")
(fun fmt t -> Format.fprintf fmt "%a" format_typ t))
2020-12-03 22:11:41 +03:00
ts
| TStruct s -> Format.fprintf fmt "%a" A.StructName.format_t s
| TEnum e -> Format.fprintf fmt "%a" A.EnumName.format_t e
| TOption t ->
Format.fprintf fmt "@[<hov 2>%a@ %s@]" format_typ_with_parens t "eoption"
| TArrow ([t1], t2) ->
Format.fprintf fmt "@[<hov 2>%a@ →@ %a@]" format_typ_with_parens t1
format_typ t2
2020-12-30 00:26:10 +03:00
| TArrow (t1, t2) ->
Format.fprintf fmt "@[<hov 2>(%a)@ →@ %a@]"
2023-01-11 13:23:08 +03:00
(Format.pp_print_list
~pp_sep:(fun fmt () -> Format.fprintf fmt ",@ ")
format_typ_with_parens)
2023-01-11 13:23:08 +03:00
t1 format_typ t2
| TArray t1 -> (
match Marked.unmark (UnionFind.get (UnionFind.find t1)) with
| TAny _ when not !Cli.debug_flag -> Format.pp_print_string fmt "collection"
| _ -> Format.fprintf fmt "@[collection@ %a@]" format_typ t1)
| TAny v ->
if !Cli.debug_flag then Format.fprintf fmt "<a%d>" (Any.hash v)
else Format.pp_print_string fmt "<any>"
exception Type_error of A.any_expr * unionfind_typ * unionfind_typ
2022-07-11 12:32:23 +03:00
type mark = { pos : Pos.t; uf : unionfind_typ }
(** Raises an error if unification cannot be performed. The position annotation
of the second [unionfind_typ] argument is propagated (unless it is [TAny]). *)
2021-01-14 02:17:24 +03:00
let rec unify
(ctx : A.decl_ctx)
(e : ('a, 'm A.mark) A.gexpr) (* used for error context *)
(t1 : unionfind_typ)
(t2 : unionfind_typ) : unit =
2021-01-14 02:17:24 +03:00
let unify = unify ctx in
(* Cli.debug_format "Unifying %a and %a" (format_typ ctx) t1 (format_typ ctx)
t2; *)
let t1_repr = UnionFind.get (UnionFind.find t1) in
let t2_repr = UnionFind.get (UnionFind.find t2) in
let raise_type_error () = raise (Type_error (A.AnyExpr e, t1, t2)) in
let () =
2022-07-11 12:32:23 +03:00
match Marked.unmark t1_repr, Marked.unmark t2_repr with
| TLit tl1, TLit tl2 -> if tl1 <> tl2 then raise_type_error ()
2022-07-11 12:32:23 +03:00
| TArrow (t11, t12), TArrow (t21, t22) ->
2022-09-16 19:15:30 +03:00
unify e t12 t22;
2023-01-11 13:23:08 +03:00
if List.length t11 = List.length t21 then List.iter2 (unify e) t11 t21
else raise_type_error ()
| TTuple ts1, TTuple ts2 ->
if List.length ts1 = List.length ts2 then List.iter2 (unify e) ts1 ts2
2022-07-11 12:32:23 +03:00
else raise_type_error ()
| TStruct s1, TStruct s2 ->
if not (A.StructName.equal s1 s2) then raise_type_error ()
| TEnum e1, TEnum e2 ->
if not (A.EnumName.equal e1 e2) then raise_type_error ()
| TOption t1, TOption t2 -> unify e t1 t2
| TArray t1', TArray t2' -> unify e t1' t2'
| TAny _, _ | _, TAny _ -> ()
| ( ( TLit _ | TArrow _ | TTuple _ | TStruct _ | TEnum _ | TOption _
| TArray _ ),
_ ) ->
raise_type_error ()
2020-12-30 03:02:04 +03:00
in
ignore
@@ UnionFind.merge
(fun t1 t2 -> match Marked.unmark t2 with TAny _ -> t1 | _ -> t2)
t1 t2
2022-07-11 12:32:23 +03:00
let handle_type_error ctx e t1 t2 =
(* TODO: if we get weird error messages, then it means that we should use the
persistent version of the union-find data structure. *)
let pos =
match e with
| A.AnyExpr e -> (
match Marked.get_mark e with Untyped { pos } | Typed { pos; _ } -> pos)
in
2022-07-11 12:32:23 +03:00
let t1_repr = UnionFind.get (UnionFind.find t1) in
let t2_repr = UnionFind.get (UnionFind.find t2) in
let t1_pos = Marked.get_mark t1_repr in
let t2_pos = Marked.get_mark t2_repr in
let unformat_typ typ =
let buf = Buffer.create 59 in
let ppf = Format.formatter_of_buffer buf in
(* set infinite width to disable line cuts *)
Format.pp_set_margin ppf max_int;
format_typ ctx ppf typ;
Format.pp_print_flush ppf ();
Buffer.contents buf
in
let t1_s fmt () =
Cli.format_with_style [ANSITerminal.yellow] fmt (unformat_typ t1)
in
let t2_s fmt () =
Cli.format_with_style [ANSITerminal.yellow] fmt (unformat_typ t2)
in
Errors.raise_multispanned_error
[
( Some
(Format.asprintf
"Error coming from typechecking the following expression:"),
pos );
2022-07-11 12:32:23 +03:00
Some (Format.asprintf "Type %a coming from expression:" t1_s ()), t1_pos;
Some (Format.asprintf "Type %a coming from expression:" t2_s ()), t2_pos;
]
"Error during typechecking, incompatible types:\n%a %a\n%a %a"
(Cli.format_with_style [ANSITerminal.blue; ANSITerminal.Bold])
"-->" t1_s ()
(Cli.format_with_style [ANSITerminal.blue; ANSITerminal.Bold])
"-->" t2_s ()
let lit_type (type a) (lit : a A.glit) : naked_typ =
match lit with
| LBool _ -> TLit TBool
| LInt _ -> TLit TInt
| LRat _ -> TLit TRat
| LMoney _ -> TLit TMoney
| LDate _ -> TLit TDate
| LDuration _ -> TLit TDuration
| LUnit -> TLit TUnit
| LEmptyError -> TAny (Any.fresh ())
(** [op_type] and [resolve_overload] are a bit similar, and work on disjoint
sets of operators. However, their assumptions are different so we keep the
functions separate. In particular [resolve_overloads] requires its argument
types to be known in advance. *)
let polymorphic_op_type (op : ('a, Operator.polymorphic) A.operator Marked.pos)
: unionfind_typ =
let open Operator in
let pos = Marked.get_mark op in
let any = lazy (UnionFind.make (TAny (Any.fresh ()), pos)) in
let any2 = lazy (UnionFind.make (TAny (Any.fresh ()), pos)) in
let bt = lazy (UnionFind.make (TLit TBool, pos)) in
let it = lazy (UnionFind.make (TLit TInt, pos)) in
let array a = lazy (UnionFind.make (TArray (Lazy.force a), pos)) in
let ( @-> ) x y =
2023-01-11 13:23:08 +03:00
lazy (UnionFind.make (TArrow (List.map Lazy.force x, Lazy.force y), pos))
in
let ty =
match Marked.unmark op with
2023-01-11 13:23:08 +03:00
| Fold -> [[any2; any] @-> any2; any2; array any] @-> any2
| Eq -> [any; any] @-> bt
| Map -> [[any] @-> any2; array any] @-> array any2
| Filter -> [[any] @-> bt; array any] @-> array any
| Reduce -> [[any; any] @-> any; any; array any] @-> any
| Concat -> [array any; array any] @-> array any
| Log (PosRecordIfTrueBool, _) -> [bt] @-> bt
| Log _ -> [any] @-> any
| Length -> [array any] @-> it
in
Lazy.force ty
let resolve_overload_ret_type
(ctx : A.decl_ctx)
e
(op : ('a A.any, Operator.overloaded) A.operator)
tys : unionfind_typ =
let op_ty =
Operator.overload_type ctx
(Marked.mark (Expr.pos e) op)
(List.map (typ_to_ast ~unsafe:true) tys)
(* We use [unsafe] because the error is caught below *)
in
ast_to_typ (Type.arrow_return op_ty)
2020-12-14 20:09:38 +03:00
(** {1 Double-directed typing} *)
module Env = struct
type 'e t = {
vars : ('e, unionfind_typ) Var.Map.t;
scope_vars : A.typ A.ScopeVar.Map.t;
scopes : A.typ A.ScopeVar.Map.t A.ScopeName.Map.t;
toplevel_vars : A.typ A.TopdefName.Map.t;
}
2020-11-23 12:44:06 +03:00
let empty =
{
vars = Var.Map.empty;
scope_vars = A.ScopeVar.Map.empty;
scopes = A.ScopeName.Map.empty;
toplevel_vars = A.TopdefName.Map.empty;
}
let get t v = Var.Map.find_opt v t.vars
let get_scope_var t sv = A.ScopeVar.Map.find_opt sv t.scope_vars
let get_toplevel_var t v = A.TopdefName.Map.find_opt v t.toplevel_vars
Make scopes directly callable Quite a few changes are included here, some of which have some extra implications visible in the language: - adds the `Scope of { -- input_v: value; ... }` construct in the language - handle it down the pipeline: * `ScopeCall` in the surface AST * `EScopeCall` in desugared and scopelang * expressions are now traversed to detect dependencies between scopes * transformed into a normal function call in dcalc - defining a scope now implicitely defines a structure with the same name, with the output variables of the scope defined as fields. This allows us to type the return value from a scope call and access its fields easily. * the implications are mostly in surface/name_resolution.ml code-wise * the `Scope_out` struct that was defined in scope_to_dcalc is no longer needed/used and the fields are no longer renamed (changes some outputs; the explicit suffix for variables with multiple states is ignored as well) * one benefit is that disambiguation works just like for structures when there are conflicts on field names * however, it's now a conflict if a scope and a structure have the same name (side-note: issues with conflicting enum / struct names or scope variables / subscope names were silent and are now properly reported) - you can consequently use scope names as types for variables as well. Writing literals is not allowed though, they can only be obtained by calling the scope. Remaining TODOs: - context variables are not handled properly at the moment - error handling on invalid calls - tests show a small error message regression; lots of examples will need tweaking to avoid scope/struct name or struct fields / output variable conflicts - add a `->` syntax to make struct field access distinct from scope output var access, enforced with typing. This is expected to reduce confusion of users and add a little typing precision. - document the new syntax & implications (tutorial, cheat-sheet) - a consequence of the changes is that subscope variables also can now be typed. A possible future evolution / simplification would be to rewrite subscopes as explicit scope calls early in the pipeline. That could also allow to manipulate them as expressions (bind them in let-ins, return them...)
2022-10-21 16:47:17 +03:00
let get_subscope_out_var t scope var =
Option.bind (A.ScopeName.Map.find_opt scope t.scopes) (fun vmap ->
A.ScopeVar.Map.find_opt var vmap)
let add v tau t = { t with vars = Var.Map.add v tau t.vars }
let add_var v typ t = add v (ast_to_typ typ) t
let add_scope_var v typ t =
{ t with scope_vars = A.ScopeVar.Map.add v typ t.scope_vars }
let add_scope scope_name ~vars t =
{ t with scopes = A.ScopeName.Map.add scope_name vars t.scopes }
let add_toplevel_var v typ t =
{ t with toplevel_vars = A.TopdefName.Map.add v typ t.toplevel_vars }
let open_scope scope_name t =
let scope_vars =
A.ScopeVar.Map.union
(fun _ _ -> assert false)
t.scope_vars
(A.ScopeName.Map.find scope_name t.scopes)
in
{ t with scope_vars }
end
let add_pos e ty = Marked.mark (Expr.pos e) ty
let ty (_, { uf; _ }) = uf
2020-12-14 20:09:38 +03:00
(** Infers the most permissive type from an expression *)
let rec typecheck_expr_bottom_up :
type a m.
A.decl_ctx ->
(a, m A.mark) A.gexpr Env.t ->
(a, m A.mark) A.gexpr ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
(a, mark) A.boxed_gexpr =
fun ctx env e ->
2022-10-07 16:59:10 +03:00
typecheck_expr_top_down ctx env
(UnionFind.make (add_pos e (TAny (Any.fresh ()))))
e
2021-01-13 14:04:14 +03:00
(** Checks whether the expression can be typed with the provided type *)
and typecheck_expr_top_down :
type a m.
A.decl_ctx ->
(a, m A.mark) A.gexpr Env.t ->
unionfind_typ ->
(a, m A.mark) A.gexpr ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
(a, mark) A.boxed_gexpr =
fun ctx env tau e ->
(* Cli.debug_format "Propagating type %a for naked_expr %a" (format_typ ctx)
tau (Expr.format ctx) e; *)
let pos_e = Expr.pos e in
let () =
(* If there already is a type annotation on the given expr, ensure it
matches *)
match Marked.get_mark e with
| A.Untyped _ | A.Typed { A.ty = A.TAny, _; _ } -> ()
| A.Typed { A.ty; _ } -> unify ctx e tau (ast_to_typ ty)
in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let context_mark = { uf = tau; pos = pos_e } in
let uf_mark uf =
(* Unify with the supplied type first, and return the mark *)
unify ctx e uf tau;
{ uf; pos = pos_e }
2022-07-11 12:34:01 +03:00
in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let unionfind ?(pos = e) t = UnionFind.make (add_pos pos t) in
let ty_mark ty = uf_mark (unionfind ty) in
match Marked.unmark e with
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
| A.ELocation loc ->
let ty_opt =
match loc with
| DesugaredScopeVar (v, _) | ScopelangScopeVar v ->
Env.get_scope_var env (Marked.unmark v)
| SubScopeVar (scope, _, v) ->
Make scopes directly callable Quite a few changes are included here, some of which have some extra implications visible in the language: - adds the `Scope of { -- input_v: value; ... }` construct in the language - handle it down the pipeline: * `ScopeCall` in the surface AST * `EScopeCall` in desugared and scopelang * expressions are now traversed to detect dependencies between scopes * transformed into a normal function call in dcalc - defining a scope now implicitely defines a structure with the same name, with the output variables of the scope defined as fields. This allows us to type the return value from a scope call and access its fields easily. * the implications are mostly in surface/name_resolution.ml code-wise * the `Scope_out` struct that was defined in scope_to_dcalc is no longer needed/used and the fields are no longer renamed (changes some outputs; the explicit suffix for variables with multiple states is ignored as well) * one benefit is that disambiguation works just like for structures when there are conflicts on field names * however, it's now a conflict if a scope and a structure have the same name (side-note: issues with conflicting enum / struct names or scope variables / subscope names were silent and are now properly reported) - you can consequently use scope names as types for variables as well. Writing literals is not allowed though, they can only be obtained by calling the scope. Remaining TODOs: - context variables are not handled properly at the moment - error handling on invalid calls - tests show a small error message regression; lots of examples will need tweaking to avoid scope/struct name or struct fields / output variable conflicts - add a `->` syntax to make struct field access distinct from scope output var access, enforced with typing. This is expected to reduce confusion of users and add a little typing precision. - document the new syntax & implications (tutorial, cheat-sheet) - a consequence of the changes is that subscope variables also can now be typed. A possible future evolution / simplification would be to rewrite subscopes as explicit scope calls early in the pipeline. That could also allow to manipulate them as expressions (bind them in let-ins, return them...)
2022-10-21 16:47:17 +03:00
Env.get_subscope_out_var env scope (Marked.unmark v)
| ToplevelVar v -> Env.get_toplevel_var env (Marked.unmark v)
in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let ty =
match ty_opt with
| Some ty -> ty
| None ->
Errors.raise_spanned_error pos_e "Reference to %a not found"
(Expr.format ctx) e
in
Expr.elocation loc (uf_mark (ast_to_typ ty))
| A.EStruct { name; fields } ->
let mark = ty_mark (TStruct name) in
let str = A.StructName.Map.find name ctx.A.ctx_structs in
let _check_fields : unit =
let missing_fields, extra_fields =
A.StructField.Map.fold
(fun fld x (remaining, extra) ->
if A.StructField.Map.mem fld remaining then
A.StructField.Map.remove fld remaining, extra
else remaining, A.StructField.Map.add fld x extra)
fields
(str, A.StructField.Map.empty)
in
let errs =
List.map
(fun (f, ty) ->
( Some (Format.asprintf "Missing field %a" A.StructField.format_t f),
Marked.get_mark ty ))
(A.StructField.Map.bindings missing_fields)
@ List.map
(fun (f, ef) ->
let dup = A.StructField.Map.mem f str in
( Some
(Format.asprintf "%s field %a"
(if dup then "Duplicate" else "Unknown")
A.StructField.format_t f),
Expr.pos ef ))
(A.StructField.Map.bindings extra_fields)
in
if errs <> [] then
Errors.raise_multispanned_error errs
"Mismatching field definitions for structure %a" A.StructName.format_t
name
in
let fields' =
A.StructField.Map.mapi
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
(fun f_name f_e ->
let f_ty = A.StructField.Map.find f_name str in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
typecheck_expr_top_down ctx env (ast_to_typ f_ty) f_e)
fields
in
Expr.estruct name fields' mark
| A.EDStructAccess { e = e_struct; name_opt; field } ->
let t_struct =
match name_opt with
| Some name -> TStruct name
| None -> TAny (Any.fresh ())
in
let e_struct' =
typecheck_expr_top_down ctx env (unionfind t_struct) e_struct
in
let name =
match UnionFind.get (ty e_struct') with
| TStruct name, _ -> name
| TAny _, _ ->
Printf.ksprintf failwith
"Disambiguation failed before reaching field %s" field
| _ ->
Errors.raise_spanned_error (Expr.pos e)
"This is not a structure, cannot access field %s (%a)" field
(format_typ ctx) (ty e_struct')
in
let fld_ty =
let str =
try A.StructName.Map.find name ctx.A.ctx_structs
with Not_found ->
Errors.raise_spanned_error pos_e "No structure %a found"
A.StructName.format_t name
in
let field =
let candidate_structs =
try A.IdentName.Map.find field ctx.ctx_struct_fields
with Not_found ->
Errors.raise_spanned_error context_mark.pos
"Field %s does not belong to structure %a (no structure defines \
it)"
field A.StructName.format_t name
in
try A.StructName.Map.find name candidate_structs
with Not_found ->
Errors.raise_spanned_error context_mark.pos
"Field %s does not belong to structure %a, but to %a" field
A.StructName.format_t name
(Format.pp_print_list
~pp_sep:(fun ppf () -> Format.fprintf ppf "@ or@ ")
A.StructName.format_t)
(List.map fst (A.StructName.Map.bindings candidate_structs))
in
A.StructField.Map.find field str
in
let mark = uf_mark (ast_to_typ fld_ty) in
Expr.edstructaccess e_struct' field (Some name) mark
| A.EStructAccess { e = e_struct; name; field } ->
let fld_ty =
let str =
try A.StructName.Map.find name ctx.A.ctx_structs
with Not_found ->
Errors.raise_spanned_error pos_e "No structure %a found"
A.StructName.format_t name
in
try A.StructField.Map.find field str
with Not_found ->
Errors.raise_multispanned_error
[
None, pos_e;
( Some "Structure %a declared here",
Marked.get_mark (A.StructName.get_info name) );
]
"Structure %a doesn't define a field %a" A.StructName.format_t name
A.StructField.format_t field
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
in
let mark = uf_mark (ast_to_typ fld_ty) in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let e_struct' =
typecheck_expr_top_down ctx env (unionfind (TStruct name)) e_struct
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
in
Expr.estructaccess e_struct' field name mark
| A.EInj { name; cons; e = e_enum } ->
let mark = uf_mark (unionfind (TEnum name)) in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let e_enum' =
2022-09-26 19:19:39 +03:00
typecheck_expr_top_down ctx env
(ast_to_typ
(A.EnumConstructor.Map.find cons
(A.EnumName.Map.find name ctx.A.ctx_enums)))
2022-09-26 19:19:39 +03:00
e_enum
in
Expr.einj e_enum' cons name mark
| A.EMatch { e = e1; name; cases } ->
let cases_ty = A.EnumName.Map.find name ctx.A.ctx_enums in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let t_ret = unionfind ~pos:e1 (TAny (Any.fresh ())) in
let mark = uf_mark t_ret in
let e1' = typecheck_expr_top_down ctx env (unionfind (TEnum name)) e1 in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let cases' =
A.EnumConstructor.Map.mapi
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
(fun c_name e ->
let c_ty = A.EnumConstructor.Map.find c_name cases_ty in
(* For now our constructors are limited to zero or one argument. If
there is a change to allow for multiple arguments, it might be
easier to use tuples directly. *)
2023-01-11 13:23:08 +03:00
let e_ty = unionfind ~pos:e (TArrow ([ast_to_typ c_ty], t_ret)) in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
typecheck_expr_top_down ctx env e_ty e)
cases
in
Expr.ematch e1' name cases' mark
| A.EScopeCall { scope; args } ->
let scope_out_struct =
(A.ScopeName.Map.find scope ctx.ctx_scopes).out_struct_name
in
Make scopes directly callable Quite a few changes are included here, some of which have some extra implications visible in the language: - adds the `Scope of { -- input_v: value; ... }` construct in the language - handle it down the pipeline: * `ScopeCall` in the surface AST * `EScopeCall` in desugared and scopelang * expressions are now traversed to detect dependencies between scopes * transformed into a normal function call in dcalc - defining a scope now implicitely defines a structure with the same name, with the output variables of the scope defined as fields. This allows us to type the return value from a scope call and access its fields easily. * the implications are mostly in surface/name_resolution.ml code-wise * the `Scope_out` struct that was defined in scope_to_dcalc is no longer needed/used and the fields are no longer renamed (changes some outputs; the explicit suffix for variables with multiple states is ignored as well) * one benefit is that disambiguation works just like for structures when there are conflicts on field names * however, it's now a conflict if a scope and a structure have the same name (side-note: issues with conflicting enum / struct names or scope variables / subscope names were silent and are now properly reported) - you can consequently use scope names as types for variables as well. Writing literals is not allowed though, they can only be obtained by calling the scope. Remaining TODOs: - context variables are not handled properly at the moment - error handling on invalid calls - tests show a small error message regression; lots of examples will need tweaking to avoid scope/struct name or struct fields / output variable conflicts - add a `->` syntax to make struct field access distinct from scope output var access, enforced with typing. This is expected to reduce confusion of users and add a little typing precision. - document the new syntax & implications (tutorial, cheat-sheet) - a consequence of the changes is that subscope variables also can now be typed. A possible future evolution / simplification would be to rewrite subscopes as explicit scope calls early in the pipeline. That could also allow to manipulate them as expressions (bind them in let-ins, return them...)
2022-10-21 16:47:17 +03:00
let mark = uf_mark (unionfind (TStruct scope_out_struct)) in
let vars = A.ScopeName.Map.find scope env.scopes in
let args' =
A.ScopeVar.Map.mapi
Make scopes directly callable Quite a few changes are included here, some of which have some extra implications visible in the language: - adds the `Scope of { -- input_v: value; ... }` construct in the language - handle it down the pipeline: * `ScopeCall` in the surface AST * `EScopeCall` in desugared and scopelang * expressions are now traversed to detect dependencies between scopes * transformed into a normal function call in dcalc - defining a scope now implicitely defines a structure with the same name, with the output variables of the scope defined as fields. This allows us to type the return value from a scope call and access its fields easily. * the implications are mostly in surface/name_resolution.ml code-wise * the `Scope_out` struct that was defined in scope_to_dcalc is no longer needed/used and the fields are no longer renamed (changes some outputs; the explicit suffix for variables with multiple states is ignored as well) * one benefit is that disambiguation works just like for structures when there are conflicts on field names * however, it's now a conflict if a scope and a structure have the same name (side-note: issues with conflicting enum / struct names or scope variables / subscope names were silent and are now properly reported) - you can consequently use scope names as types for variables as well. Writing literals is not allowed though, they can only be obtained by calling the scope. Remaining TODOs: - context variables are not handled properly at the moment - error handling on invalid calls - tests show a small error message regression; lots of examples will need tweaking to avoid scope/struct name or struct fields / output variable conflicts - add a `->` syntax to make struct field access distinct from scope output var access, enforced with typing. This is expected to reduce confusion of users and add a little typing precision. - document the new syntax & implications (tutorial, cheat-sheet) - a consequence of the changes is that subscope variables also can now be typed. A possible future evolution / simplification would be to rewrite subscopes as explicit scope calls early in the pipeline. That could also allow to manipulate them as expressions (bind them in let-ins, return them...)
2022-10-21 16:47:17 +03:00
(fun name ->
typecheck_expr_top_down ctx env
(ast_to_typ (A.ScopeVar.Map.find name vars)))
args
Make scopes directly callable Quite a few changes are included here, some of which have some extra implications visible in the language: - adds the `Scope of { -- input_v: value; ... }` construct in the language - handle it down the pipeline: * `ScopeCall` in the surface AST * `EScopeCall` in desugared and scopelang * expressions are now traversed to detect dependencies between scopes * transformed into a normal function call in dcalc - defining a scope now implicitely defines a structure with the same name, with the output variables of the scope defined as fields. This allows us to type the return value from a scope call and access its fields easily. * the implications are mostly in surface/name_resolution.ml code-wise * the `Scope_out` struct that was defined in scope_to_dcalc is no longer needed/used and the fields are no longer renamed (changes some outputs; the explicit suffix for variables with multiple states is ignored as well) * one benefit is that disambiguation works just like for structures when there are conflicts on field names * however, it's now a conflict if a scope and a structure have the same name (side-note: issues with conflicting enum / struct names or scope variables / subscope names were silent and are now properly reported) - you can consequently use scope names as types for variables as well. Writing literals is not allowed though, they can only be obtained by calling the scope. Remaining TODOs: - context variables are not handled properly at the moment - error handling on invalid calls - tests show a small error message regression; lots of examples will need tweaking to avoid scope/struct name or struct fields / output variable conflicts - add a `->` syntax to make struct field access distinct from scope output var access, enforced with typing. This is expected to reduce confusion of users and add a little typing precision. - document the new syntax & implications (tutorial, cheat-sheet) - a consequence of the changes is that subscope variables also can now be typed. A possible future evolution / simplification would be to rewrite subscopes as explicit scope calls early in the pipeline. That could also allow to manipulate them as expressions (bind them in let-ins, return them...)
2022-10-21 16:47:17 +03:00
in
Expr.escopecall scope args' mark
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
| A.ERaise ex -> Expr.eraise ex context_mark
| A.ECatch { body; exn; handler } ->
let body' = typecheck_expr_top_down ctx env tau body in
let handler' = typecheck_expr_top_down ctx env tau handler in
Expr.ecatch body' exn handler' context_mark
| A.EVar v ->
let tau' =
match Env.get env v with
| Some t -> t
| None ->
Errors.raise_spanned_error pos_e
"Variable %s not found in the current context" (Bindlib.name_of v)
in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
Expr.evar (Var.translate v) (uf_mark tau')
| A.ELit lit -> Expr.elit lit (ty_mark (lit_type lit))
| A.ETuple es ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let tys = List.map (fun _ -> unionfind (TAny (Any.fresh ()))) es in
let mark = uf_mark (unionfind (TTuple tys)) in
let es' = List.map2 (typecheck_expr_top_down ctx env) tys es in
Expr.etuple es' mark
| A.ETupleAccess { e = e1; index; size } ->
if index >= size then
Errors.raise_spanned_error (Expr.pos e)
"Tuple access out of bounds (%d/%d)" index size;
let tuple_ty =
TTuple
(List.init size (fun n ->
if n = index then tau else unionfind ~pos:e1 (TAny (Any.fresh ()))))
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
in
let e1' = typecheck_expr_top_down ctx env (unionfind ~pos:e1 tuple_ty) e1 in
Expr.etupleaccess e1' index size context_mark
| A.EAbs { binder; tys = t_args } ->
2022-06-23 15:04:51 +03:00
if Bindlib.mbinder_arity binder <> List.length t_args then
Errors.raise_spanned_error (Expr.pos e)
"function has %d variables but was supplied %d types"
2022-06-23 15:04:51 +03:00
(Bindlib.mbinder_arity binder)
(List.length t_args)
else
let tau_args = List.map ast_to_typ t_args in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let t_ret = unionfind (TAny (Any.fresh ())) in
2023-01-11 13:23:08 +03:00
let t_func = unionfind (TArrow (tau_args, t_ret)) in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let mark = uf_mark t_func in
assert (List.for_all all_resolved tau_args);
2022-06-23 15:04:51 +03:00
let xs, body = Bindlib.unmbind binder in
let xs' = Array.map Var.translate xs in
2022-06-23 15:04:51 +03:00
let env =
List.fold_left2
(fun env x tau_arg -> Env.add x tau_arg env)
env (Array.to_list xs) tau_args
in
let body' = typecheck_expr_top_down ctx env t_ret body in
let binder' = Bindlib.bind_mvar xs' (Expr.Box.lift body') in
Expr.eabs binder' (List.map typ_to_ast tau_args) mark
| A.EApp { f = (EOp { op; tys }, _) as e1; args } ->
let t_args = List.map ast_to_typ tys in
2023-01-11 13:23:08 +03:00
let t_func = unionfind (TArrow (t_args, tau)) in
let e1', args' =
Operator.kind_dispatch op
~polymorphic:(fun _ ->
(* Type the operator first, then right-to-left: polymorphic operators
are required to allow the resolution of all type variables this
way *)
let e1' = typecheck_expr_top_down ctx env t_func e1 in
let args' =
List.rev_map2
(typecheck_expr_top_down ctx env)
(List.rev t_args) (List.rev args)
in
e1', args')
~overloaded:(fun _ ->
(* Typing the arguments first is required to resolve the operator *)
let args' = List.map2 (typecheck_expr_top_down ctx env) t_args args in
let e1' = typecheck_expr_top_down ctx env t_func e1 in
e1', args')
~monomorphic:(fun _ ->
(* Here it doesn't matter but may affect the error messages *)
let e1' = typecheck_expr_top_down ctx env t_func e1 in
let args' = List.map2 (typecheck_expr_top_down ctx env) t_args args in
e1', args')
~resolved:(fun _ ->
(* This case should not fail *)
let e1' = typecheck_expr_top_down ctx env t_func e1 in
let args' = List.map2 (typecheck_expr_top_down ctx env) t_args args in
e1', args')
in
Expr.eapp e1' args' context_mark
| A.EApp { f = e1; args } ->
(* Here we type the arguments first (in order), to ensure we know the types
of the arguments if [f] is [EAbs] before disambiguation. This is also the
right order for the [let-in] form. *)
let t_args = List.map (fun _ -> unionfind (TAny (Any.fresh ()))) args in
2023-01-11 13:23:08 +03:00
let t_func = unionfind (TArrow (t_args, tau)) in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let args' = List.map2 (typecheck_expr_top_down ctx env) t_args args in
let e1' = typecheck_expr_top_down ctx env t_func e1 in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
Expr.eapp e1' args' context_mark
| A.EOp { op; tys } ->
let tys' = List.map ast_to_typ tys in
let t_ret = unionfind (TAny (Any.fresh ())) in
2023-01-11 13:23:08 +03:00
let t_func = unionfind (TArrow (tys', t_ret)) in
unify ctx e t_func tau;
let tys, mark =
Operator.kind_dispatch op
~polymorphic:(fun op ->
tys, uf_mark (polymorphic_op_type (Marked.mark pos_e op)))
~monomorphic:(fun op ->
let mark =
uf_mark
(ast_to_typ (Operator.monomorphic_type (Marked.mark pos_e op)))
in
List.map typ_to_ast tys', mark)
~overloaded:(fun op ->
unify ctx e t_ret (resolve_overload_ret_type ctx e op tys');
List.map typ_to_ast tys', { uf = t_func; pos = pos_e })
~resolved:(fun op ->
let mark =
uf_mark (ast_to_typ (Operator.resolved_type (Marked.mark pos_e op)))
in
List.map typ_to_ast tys', mark)
in
Expr.eop op tys mark
| A.EDefault { excepts; just; cons } ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let cons' = typecheck_expr_top_down ctx env tau cons in
let just' =
typecheck_expr_top_down ctx env (unionfind ~pos:just (TLit TBool)) just
in
let excepts' = List.map (typecheck_expr_top_down ctx env tau) excepts in
Expr.edefault excepts' just' cons' context_mark
| A.EIfThenElse { cond; etrue = et; efalse = ef } ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let et' = typecheck_expr_top_down ctx env tau et in
let ef' = typecheck_expr_top_down ctx env tau ef in
let cond' =
typecheck_expr_top_down ctx env (unionfind ~pos:cond (TLit TBool)) cond
in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
Expr.eifthenelse cond' et' ef' context_mark
| A.EAssert e1 ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let mark = uf_mark (unionfind (TLit TUnit)) in
let e1' =
typecheck_expr_top_down ctx env (unionfind ~pos:e1 (TLit TBool)) e1
2022-07-11 12:32:23 +03:00
in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
Expr.eassert e1' mark
| A.EErrorOnEmpty e1 ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let e1' = typecheck_expr_top_down ctx env tau e1 in
Expr.eerroronempty e1' context_mark
| A.EArray es ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let cell_type = unionfind (TAny (Any.fresh ())) in
let mark = uf_mark (unionfind (TArray cell_type)) in
let es' = List.map (typecheck_expr_top_down ctx env cell_type) es in
Expr.earray es' mark
2022-07-11 12:32:23 +03:00
let wrap ctx f e =
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
try f e
2022-07-11 12:32:23 +03:00
with Type_error (e, ty1, ty2) -> (
let bt = Printexc.get_raw_backtrace () in
try handle_type_error ctx e ty1 ty2
with e -> Printexc.raise_with_backtrace e bt)
2020-11-23 12:44:06 +03:00
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let wrap_expr ctx f e =
(* We need to unbox here, because the typing may otherwise be stored in
Bindlib closures and not yet applied, and would escape the `try..with` *)
wrap ctx (fun e -> Expr.unbox (f e)) e
2020-12-14 20:09:38 +03:00
(** {1 API} *)
let get_ty_mark { uf; pos } = A.Typed { ty = typ_to_ast uf; pos }
let expr_raw
(type a)
(ctx : A.decl_ctx)
?(env = Env.empty)
?(typ : A.typ option)
(e : (a, 'm) A.gexpr) : (a, mark) A.gexpr =
let fty =
match typ with
| None -> typecheck_expr_bottom_up ctx env
| Some typ -> typecheck_expr_top_down ctx env (ast_to_typ typ)
in
wrap_expr ctx fty e
let check_expr ctx ?env ?typ e =
Expr.map_marks
~f:(fun { pos; _ } -> A.Untyped { pos })
(expr_raw ctx ?env ?typ e)
(* Infer the type of an expression *)
let expr ctx ?env ?typ e =
Expr.map_marks ~f:get_ty_mark (expr_raw ctx ?env ?typ e)
let rec scope_body_expr ctx env ty_out body_expr =
match body_expr with
| A.Result e ->
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let e' = wrap_expr ctx (typecheck_expr_top_down ctx env ty_out) e in
let e' = Expr.map_marks ~f:get_ty_mark e' in
Bindlib.box_apply (fun e -> A.Result e) (Expr.Box.lift e')
| A.ScopeLet
{
scope_let_kind;
scope_let_typ;
scope_let_expr = e0;
scope_let_next;
scope_let_pos;
} ->
let ty_e = ast_to_typ scope_let_typ in
Swap boxing and annotations in expressions This was the only reasonable solution I found to the issue raised [here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884). This was a pretty tedious rewrite, but it should now ensure we are doing things correctly. As a bonus, the "smart" expression constructors are now used everywhere to build expressions (so another refactoring like this one should be much easier) and this makes the code overall feel more straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!) --- Basically, we were using values of type `gexpr box = naked_gexpr marked box` throughout when (re-)building expressions. This was done 99% of the time by using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In lots of places, we needed to recover the annotation of this expression later on, typically to build its parent term (to inherit the position, or build the type). Since it wasn't always possible to wrap these uses within `box_apply` (esp. as bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`, just to recover the position or type. This had the very unpleasant effect of forcing the resolution of the whole box (including applying any stored closures) to reach the top-level annotation which isn't even dependant on specific variable bindings. Then, generally, throwing away the result. Therefore, the change proposed here transforms - `naked_gexpr marked Bindlib.box` into - `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for convenience) This means only 1. not fitting the mark into the box right away when building, and 2. accessing the top-level mark directly without unboxing The functions for building terms from module `Shared_ast.Expr` could be changed easily. But then they needed to be consistently used throughout, without manually building terms through `Bindlib.apply_box` -- which covers most of the changes in this patch. `Expr.Box.inj` is provided to swap back to a box, before binding for example. Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`, which hints at the amount of unnecessary work we were doing --'
2022-10-06 20:13:45 +03:00
let e = wrap_expr ctx (typecheck_expr_bottom_up ctx env) e0 in
wrap ctx (fun t -> unify ctx e0 (ty e) t) ty_e;
(* We could use [typecheck_expr_top_down] rather than this manual
unification, but we get better messages with this order of the [unify]
parameters, which keeps location of the type as defined instead of as
inferred. *)
let var, next = Bindlib.unbind scope_let_next in
let env = Env.add var ty_e env in
let next = scope_body_expr ctx env ty_out next in
let scope_let_next = Bindlib.bind_var (Var.translate var) next in
Bindlib.box_apply2
(fun scope_let_expr scope_let_next ->
A.ScopeLet
{
scope_let_kind;
scope_let_typ;
scope_let_expr;
scope_let_next;
scope_let_pos;
})
(Expr.Box.lift (Expr.map_marks ~f:get_ty_mark e))
scope_let_next
let scope_body ctx env body =
let get_pos struct_name =
Marked.get_mark (A.StructName.get_info struct_name)
in
let struct_ty struct_name =
UnionFind.make (Marked.mark (get_pos struct_name) (TStruct struct_name))
in
let ty_in = struct_ty body.A.scope_body_input_struct in
let ty_out = struct_ty body.A.scope_body_output_struct in
let var, e = Bindlib.unbind body.A.scope_body_expr in
let env = Env.add var ty_in env in
let e' = scope_body_expr ctx env ty_out e in
( Bindlib.box_apply
(fun scope_body_expr -> { body with scope_body_expr })
(Bindlib.bind_var (Var.translate var) e'),
UnionFind.make
(Marked.mark
(get_pos body.A.scope_body_output_struct)
2023-01-11 13:23:08 +03:00
(TArrow ([ty_in], ty_out))) )
let rec scopes ctx env = function
| A.Nil -> Bindlib.box A.Nil
| A.Cons (item, next_bind) ->
let var, next = Bindlib.unbind next_bind in
let env, def =
match item with
| A.ScopeDef (name, body) ->
let body_e, ty_scope = scope_body ctx env body in
( Env.add var ty_scope env,
Bindlib.box_apply (fun body -> A.ScopeDef (name, body)) body_e )
| A.Topdef (name, typ, e) ->
let e' = expr_raw ctx ~env ~typ e in
let uf = (Marked.get_mark e').uf in
let e' = Expr.map_marks ~f:get_ty_mark e' in
( Env.add var uf env,
Bindlib.box_apply
(fun e -> A.Topdef (name, typ, e))
(Expr.Box.lift e') )
in
let next' = scopes ctx env next in
let next_bind' = Bindlib.bind_var (Var.translate var) next' in
Bindlib.box_apply2 (fun item next -> A.Cons (item, next)) def next_bind'
let program prg =
let code_items =
Bindlib.unbox (scopes prg.A.decl_ctx Env.empty prg.A.code_items)
in
{ prg with code_items }