Remove `assert` as a reserved keyword by aspeddro · Pull Request #8399 · rescript-lang/rescript

aspeddro · 2026-04-29T14:40:14Z

Summary

Removes the Assert token from the scanner/tokenizer, its parse case from the parser (res_core.ml), and the Assert entries from grammar predicates (is_expr_start, is_block_expr_start)
Adds external assert: bool => 'a = "%assert" to Pervasives.res so the function is available everywhere without an explicit import (Pervasives is auto-opened for all runtime files)
Handles %assert specially in the type checker (typecore.ml) to produce Texp_assert with the same type rules as the former Pexp_assert node: unit in statement position, polymorphic 'a when the argument is false — preventing warning 21 from firing on assert(expr) used as a statement
Adds a matching special case in translcore.ml so the existing assert_failed helper (which uses the call-site location) generates the correct JS output with source location in the thrown Assert_failure exception
Rejected first-class assert usage, such as let f = assert, because it cannot preserve -noassert semantics without evaluating arguments first.

Generated JS output

// assert(x == 1) at line 5, col 2
throw {
  RE_EXN_ID: "Assert_failure",
  _1: ["filename.res", 5, 2],
  Error: new Error()
};

Test plan

make test-syntax — syntax parser and printer tests pass
make lib — all runtime modules compile (181 modules)
make test — full test suite

🤖 Generated with Claude Code and Codex

`assert` is now a regular function call `assert(expr)` rather than a keyword expression. This removes the `Assert` token from the scanner, its parse case from the parser, and the corresponding grammar predicates. To preserve correct semantics and source-location information in thrown `Assert_failure` exceptions, the `%assert` primitive is handled specially in the type checker (`typecore.ml`): it produces `Texp_assert` with the same type rules as the former `Pexp_assert` node (`unit` in statement position, `'a` when the argument is `false`). `translcore.ml` has a matching special case so the existing `assert_failed` helper — which uses the call-site location — generates the correct JS output. `Pervasives.res` now declares `external assert: bool => 'a = "%assert"` so the function is available everywhere without an explicit import. Signed-Off-By: Pedro Castro <aspeddro@gmail.com> Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

aspeddro · 2026-04-29T14:40:55Z

I should remove Pexp_assert of expression from parsetree?

pkg-pr-new · 2026-04-29T16:21:32Z

Open in StackBlitz

rescript

npm i https://pkg.pr.new/rescript@8399

@rescript/darwin-arm64

npm i https://pkg.pr.new/@rescript/darwin-arm64@8399

@rescript/darwin-x64

npm i https://pkg.pr.new/@rescript/darwin-x64@8399

@rescript/linux-arm64

npm i https://pkg.pr.new/@rescript/linux-arm64@8399

@rescript/linux-x64

npm i https://pkg.pr.new/@rescript/linux-x64@8399

@rescript/runtime

npm i https://pkg.pr.new/@rescript/runtime@8399

@rescript/win32-x64

npm i https://pkg.pr.new/@rescript/win32-x64@8399

commit: 381abce

cknitt · 2026-04-30T07:59:05Z

I should remove Pexp_assert of expression from parsetree?

Yes, if it is unused now, it should be removed from Parsetree.

cknitt · 2026-04-30T08:16:07Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2c716366e2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-30T08:27:23Z

+  | Pexp_apply {funct = {pexp_desc = Pexp_ident lid}; args = [(Nolabel, scond)]}
+    when match Env.lookup_value lid.txt env with
+         | _, {val_kind = Val_prim {Primitive.prim_name = "%assert"}} -> true
+         | _ -> false
+         | exception Not_found -> false ->


Handle %assert when used via aliases or higher-order values

This special-case only rewrites direct assert(cond) calls, so %assert can still escape as a first-class value (e.g. let f = assert; f(true)). In that path the backend no longer sees Texp_assert, and %assert falls through primitive lowering paths that are not implemented for it, which can surface as an internal primitive-conversion failure instead of normal assert semantics. Please either reject non-direct %assert usage in typing or add full primitive lowering support for %assert as a value.

Useful? React with 👍 / 👎.

Fixed in adbb76c

Reverted in 381abce

When assert is used as a value (e.g. let f = assert), transl_primitive now generates a proper wrapper function instead of falling through to Pccall, which would produce a call to a non-existent runtime function. The generated lambda is: fun assert_cond -> if assert_cond then () else raise Assert_failure(file, line, col), using the location where assert appears in source. This is the best possible location for an alias since the call-site location is not available at primitive translation time. assert_failed_at is extracted as a location-based helper alongside the existing assert_failed (which takes a typed expression). Signed-Off-By: Pedro Castro <aspeddro@gmail.com> Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

cknitt · 2026-04-30T15:07:09Z

    (* let module M = ME in E *)
  | Pexp_letexception of extension_constructor * expression
    (* let exception C in E *)
-  | Pexp_assert of expression


Parsetree0 represents the old AST and needs to remain unchanged for PPX to continue to work.

So we can only remove Pexp_assert from Parsetree, not Parsetree0.

I tried removing it from parsetree, but Pexp_assert is used in parsetree0, ast_mapper_to0.ml and other modules.

I undid the commit.

cknitt · 2026-05-03T11:42:41Z

@aspeddro I pushed a commit that removes Pexp_assert from Parsetree, but not Parsetree0.

In local review, Codex found another issue though:

The direct-call typechecker special case gives assert(false) the expected type and every other direct assert(...) call unit, but the exported value itself is bool => 'a.

As soon as assert is aliased or passed first-class, non-false calls can be accepted at arbitrary result types even though the eta-expanded primitive returns unit on success.

For example:

let f = assert
let x: int = f(true)

This can typecheck but produces a unit/undefined value at runtime.

Consider exposing assert as bool => unit and keeping the existing direct-call special case for literal false.

aspeddro · 2026-05-13T20:24:52Z

@cknitt changes done. Ready for review!!

fhammerschmidt · 2026-05-22T12:05:19Z

@codex review once more

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0cbe0e833f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-05-22T12:12:49Z

+  if p.Primitive.prim_name = "%assert" then
+    let param = Ident.create "assert_cond" in
+    Lfunction
+      {
+        params = [param];
+        attr = default_function_attribute;
+        loc;
+        body =
+          (if !Clflags.noassert then lambda_unit
+           else Lifthenelse (Lvar param, lambda_unit, assert_failed_at loc));
+      }


Preserve -noassert behavior for first-class assert

When %assert is translated as a normal one-argument function here, calls through an alias still evaluate the argument even under -noassert (e.g. let f = assert; f(sideEffect())), because function arguments are evaluated before the call. That diverges from assertion-disabled semantics used elsewhere in this commit (direct assert(...) is lowered to () without evaluating the condition), so enabling -noassert no longer reliably removes assertion-condition side effects once assert is passed around as a value.

Useful? React with 👍 / 👎.

To preserve -noassert behavior %assert should be a non-first-class again.

To make first-class assert correct, we would need a much deeper special representation: track aliases of %assert, recognize applications through those aliases, possibly through records/modules/higher-order arguments, and lower the entire application before evaluating the condition.

We can add some limitation, assert can only be used as a direct call like assert(condition). It cannot be aliased or passed as a function.

What you think @cknitt @fhammerschmidt ?

@aspeddro Sounds like a good compromise - can you give it a try?

See 381abce

Signed-off-by: Pedro Castro <aspeddro@gmail.com>

cknitt · 2026-05-29T07:42:32Z

@codex review

chatgpt-codex-connector · 2026-05-29T07:47:16Z

Codex Review: Didn't find any major issues. Bravo.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

update syntax_tests

2c71636

chatgpt-codex-connector Bot reviewed Apr 30, 2026

View reviewed changes

cknitt reviewed Apr 30, 2026

View reviewed changes

aspeddro force-pushed the remove-assert-keyword branch from 1ce54a0 to adbb76c Compare April 30, 2026 17:40

aspeddro and others added 2 commits May 1, 2026 13:42

update CHANGELOG.md

3548f36

Remove Pexp_assert from Parsetree

d22712c

aspeddro added 2 commits May 3, 2026 18:17

Change return type to unit and add docstrings

842e05d

Merge branch 'master' into remove-assert-keyword

14b5423

aspeddro requested a review from cknitt May 13, 2026 20:24

restore changelog

0cbe0e8

chatgpt-codex-connector Bot reviewed May 22, 2026

View reviewed changes

Reject first-class assert usage

381abce

Signed-off-by: Pedro Castro <aspeddro@gmail.com>

Conversation

aspeddro commented Apr 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Generated JS output

Test plan

Uh oh!

aspeddro commented Apr 29, 2026

Uh oh!

pkg-pr-new Bot commented Apr 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cknitt commented Apr 30, 2026

Uh oh!

cknitt commented Apr 30, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

aspeddro Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

aspeddro May 26, 2026

Choose a reason for hiding this comment

Uh oh!

cknitt Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

aspeddro Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cknitt commented May 3, 2026

Uh oh!

aspeddro commented May 13, 2026

Uh oh!

fhammerschmidt commented May 22, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot May 22, 2026

Choose a reason for hiding this comment

Uh oh!

aspeddro May 25, 2026

Choose a reason for hiding this comment

Uh oh!

cknitt May 26, 2026

Choose a reason for hiding this comment

Uh oh!

aspeddro May 26, 2026

Choose a reason for hiding this comment

Uh oh!

cknitt commented May 29, 2026

Uh oh!

chatgpt-codex-connector Bot commented May 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

aspeddro commented Apr 29, 2026 •

edited

Loading

pkg-pr-new Bot commented Apr 29, 2026 •

edited

Loading

aspeddro Apr 30, 2026 •

edited

Loading