Skip to content

Review of OCaml-derived language features #9074

Open
4 of 11 issues completed
Open

Description

@cknitt

ReScript inherited a number of language features from OCaml that most code never touches: GADTs, existential types, locally abstract types, polymorphic annotations, functors, recursive and first-class modules, extensible variants, private types, constraint, variance annotations, destructive substitution, module type of and #...t patterns. Several of them had little or no test coverage. This issue tracks a review of these features and the bugs it found.

Method

  • Usage: counted each feature in 21 public ReScript repositories (about 550k lines, including rescript-lang.org, rescript-relay, sury, rescript-nodejs and the two large Juspay apps), in the stdlib and in our test suites.
  • Behaviour: wrote small programs for each feature and checked what they do at run time, their error messages, the formatter, and editor hover, completion and interface generation.

Findings

Run-time behaviour was correct throughout, including GADTs with @as, @tag and @unboxed, existentials packed with first-class modules, and higher-order and generative functors. Every bug was in diagnostics, the formatter or the editor tooling.

Usage outside our own test fixtures:

Feature Use
GADTs, locally abstract types rare (rescript-urql, a few bindings)
Polymorphic annotations ('a.) used in sury
constraint rescript-nodejs (Stream, Net)
open! rescript-lang.org, rescript-pages, rescript-nodejs
module type of include module type of in a .resi (rescript-urql)
#...t patterns rescript-lang.org playground
User-defined open types none; only type exn +=
module rec, with type := / with module := none
Functors, first-class modules, private types, coercion common

None of these is a removal candidate. module rec and with type := / with module := have no uses in the corpus, but the corpus is mostly libraries and misses private application code. module rec is the only way to write mutually recursive modules. with type := / with module := amounts to a few branches in the type checker and parser, so removing it would save little. The remaining features are either used or depend on machinery that stays anyway (for example, exn is itself an open type).

Tests

Bugs

Known limitations of the fixes

Not bugs, noted

  • Refutation cases (| C => .) can't be written. This came up in Empty variant type #6987 together with empty variants; that issue was closed as stale without a decision.
  • exception X followed by a line starting with ( parses as exception X(...), the same newline sensitivity as f\n(a).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions