Use SingletonStrT in case of const literal declarations

[agentcooper]

// @flow

const x = 'X';
const y: typeof x = 'Y';

Before: no errors.

After:

Cannot assign 'Y' to y because string [1] is incompatible with typeof x [2].

        1│ // @flow
        2│
        3│ const x = 'X';
 [2][1] 4│ const y: typeof x = 'Y';
        5│

Fixes #2639.

This is my first attempt to dive into Flow codebase, not sure if the implementation is sane. I would like to get some comments on this. Will add tests if everything is ok.

[agentcooper]

@samwgoldman @mroch @gabelevi any comments?

[samwgoldman]

Sorry for the radio silence on this. I really like this idea! Going to import this and have the rest of the team weigh in.

[facebook-github-bot]

@samwgoldman has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[agentcooper]

any updates on this?

[peter-leonov]

@samwgoldman 🙏 please, implement this feature! It's so of a principle of least astonishment :)

[francorisso]

Please please make this happen

[mrkev]

Oh hey, I looked a bit into what happened. Looks like this interfered with React default props for some reason, and then it fell off the radar (the flow team is busy). I'll try to find out if this is still an issue, I really like this idea.

Mind fixing the merge conflict though @agentcooper?

[agentcooper]

@mrkev I'll try to update the code over the weekend

[mrkev]

sounds good!

[agentcooper]

@mrkev 👀

[facebook-github-bot]

@mrkev has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[agentcooper]

@mrkev any updates on this?

I'm trying to rebase my diff again, but I can't solve one compiler error:

File "src/typing/statement.ml", line 2499, characters 45-55:
Error: This function has type Context.t -> Loc.t Ast.Expression.t -> Type.t
       It is applied to too many arguments; maybe you forgot a `;'.

and line 2499 is:

              | VariableDeclaration.Const -> expression ~is_const:true cx expr

Here is the full patch I'm trying to apply:

diff --git a/newtests/gen_flow_files_command/test.js b/newtests/gen_flow_files_command/test.js
index da5be6130..b373edfee 100644
--- a/newtests/gen_flow_files_command/test.js
+++ b/newtests/gen_flow_files_command/test.js
@@ -146,7 +146,7 @@ export default suite(({addFile, addFiles, flowCmd}) => [
           // @flow
 
           declare export default number;
-          declare export var str: string;
+          declare export var str: "asdf";
 
 
         `,
diff --git a/src/typing/statement.ml b/src/typing/statement.ml
index e210a296a..8f3582ce3 100644
--- a/src/typing/statement.ml
+++ b/src/typing/statement.ml
@@ -2494,7 +2494,11 @@ and variable cx kind ?if_uninitialized (_, vdecl) = Ast.Statement.(
         Type_inference_hooks_js.(dispatch_lval_hook cx name loc (Val t));
         (match init with
           | Some ((rhs_loc, _) as expr) ->
-            let rhs = expression cx expr in
+            let rhs =
+              match kind with
+              | VariableDeclaration.Const -> expression ~is_const:true cx expr
+              | _ -> expression cx expr
+            in
             (**
              * Const and let variables are not declared during evaluation of
              * their initializer expressions.
@@ -2596,8 +2600,8 @@ and mixin_element_spread cx (loc, e) =
     Flow.flow_t cx (arr, DefT (reason, ArrT (ArrayAT(tvar, None))));
   )
 
-and expression ?(is_cond=false) cx (loc, e) : Type.t * unit Ast.Expression.t' =
-  let t, e' = expression_ ~is_cond cx loc e in
+and expression ?(is_cond=false) ?(is_const=false) cx (loc, e) : Type.t * unit Ast.Expression.t' =
+  let t, e' = expression_ ~is_cond ~is_const cx loc e in
   Env.add_type_table cx loc t;
   t, e'
 
@@ -2610,11 +2614,11 @@ and this_ cx loc = Ast.Expression.(
 and super_ cx loc =
   Env.var_ref cx (internal_name "super") loc
 
-and expression_ ~is_cond cx loc e : Type.t * unit Ast.Expression.t' =
+and expression_ ~is_cond ~is_const cx loc e : Type.t * unit Ast.Expression.t' =
   let ex = (loc, e) in Ast.Expression.(match e with
 
   | Ast.Expression.Literal lit ->
-      literal cx loc lit, Ast.Expression.Literal lit
+      literal ~is_const cx loc lit, Ast.Expression.Literal lit
 
   (* Treat the identifier `undefined` as an annotation for error reporting
    * purposes. Like we do with other literals. Otherwise we end up pointing to
@@ -3935,7 +3939,7 @@ and identifier cx name loc =
   t
 
 (* traverse a literal expression, return result type *)
-and literal cx loc lit = Ast.Literal.(match lit.Ast.Literal.value with
+and literal ?(is_const=false) cx loc lit = Ast.Literal.(match lit.Ast.Literal.value with
   | String s ->
       (* It's too expensive to track literal information for large strings.*)
       let max_literal_length = Context.max_literal_length cx in
@@ -3944,7 +3948,9 @@ and literal cx loc lit = Ast.Literal.(match lit.Ast.Literal.value with
         then Literal (None, s)
         else AnyLiteral
       in
-      DefT (annot_reason (mk_reason RString loc), StrT lit)
+      if is_const
+      then DefT (annot_reason (mk_reason RString loc), SingletonStrT s)
+      else DefT (annot_reason (mk_reason RString loc), StrT lit)
 
   | Boolean b ->
       DefT (annot_reason (mk_reason RBoolean loc), BoolT (Some b))

[benwiley4000]

Hm, any status on this? Would be really nice to have - removes tons of needless (redundant) annotations in constant export files (as discussed here: #4279 (comment))

[mrkev]

Welp sorry, got busy with other projects. IIRC this was breaking some stuff with object literals (object literal strings were thought to be constant? So {a: 'foo'} was the inferred type for const a = {a: 'foo'}?). I might be wrong sorry, it was a while ago 😅

[goodmind]

@mrkev any info about this PR?

[mrkev]

Have there been any updates? I remember this broke the tests. 😅

[peter-leonov]

Just two years and ten months so far, too early to fix tests 😈

[goodmind]

I mean if author still wants to pursue this PR, Flow team is more willing to accept PRs lately, than 2 years ago

[agentcooper]

Sorry, folks, but I'm not interested in working on this anymore. I've switched to TypeScript since then. My contribution experience there was much better.

Feel free to re-use the code for another PR.

[goodmind]

Thank you! I re-opened this PR #7607

[mrkev]

Closing this one in that case 👍