feat(js): Variant, Switch, arrays, closures and references under Node #694

Merged

argus merged 2 commits from loop/684-js-slice-2 into main 2026-10-30 15:22:50 +00:00

Conversation 4 Commits 2 Files changed 11 +560 -114

orion commented 2026-10-30 09:29:53 +00:00

Collaborator

Copy link

Summary

Slice 2 of the JavaScript back end (compiler only: no syntax change, no new native, no std name the compiler looks up).

• Construct, Tag, Switch over variants: a bare variant is a number, any other variant is one class per case (L<id>_<case>, t first).
• FixedArray as an Array (literal, ArrayNew, static Items), with the checked element read/write and the language's own bounds-check panic text.
• Read/Write through Field, TupleField, Variant and Element steps; Contents is reported as unsupported (the lowering never writes it).
• Closure with Value and Boxed captures (boxes stay shared), CallClosure, adapters for functions without an environment.
• var parameters and Argument.place as reference objects (torb_ref_slot/field/element); a var parameter read as a value goes through .v.
• torb_process_exit and the TextCompare intrinsic; Copy/borrowed aggregates are cloned.
• js.allow grows by ten existing programs (matching, match-order, records, array-index-bounds, closures, closure-counts, closure-frame, places, place-counted, rest-patterns) and one new program, js-variants-arrays.trb (payload variants, element/field places, text comparison, closure over an Array). Nothing removed. Design slice table row 2 is marked done, with what it leaves out.

Evidence

Local only so far: the compiler source run under the nightly VM built each program, Node ran it, and stdout/exit/stderr matched the expectation files for all eleven programs above. CI results are added below as a comment.

Not verified

• The C and VM legs for the new js-variants-arrays.trb: .expected comes from the nightly VM and I checked it by hand, but it has no .allocations budget yet (needs a native build; to be taken from CI).
• sh tools/gates.sh a and the forge conformance-js job: not run locally (the compiler binary cannot be built here).
• Not covered on purpose: Buffer as a typed array (the IR builds it as new List, part of the slice-4 ListStorage), BoxNew, LazyForce, Place captures and Contents steps (the C back end does not support them either; they are reported as unsupported). FixedArray is a plain Array, not a typed array. Programs that print lists/maps or use callWitness (counted-closures, most array-*) stay off js.allow: they need slice 3/4.

Part of #63
Closes torbscript/language#684

## Summary Slice 2 of the JavaScript back end (compiler only: no syntax change, no new native, no std name the compiler looks up). - `Construct`, `Tag`, `Switch` over variants: a bare variant is a number, any other variant is one class per case (`L<id>_<case>`, `t` first). - `FixedArray` as an `Array` (literal, `ArrayNew`, static `Items`), with the checked element read/write and the language's own bounds-check panic text. - `Read`/`Write` through `Field`, `TupleField`, `Variant` and `Element` steps; `Contents` is reported as unsupported (the lowering never writes it). - `Closure` with `Value` and `Boxed` captures (boxes stay shared), `CallClosure`, adapters for functions without an environment. - `var` parameters and `Argument.place` as reference objects (`torb_ref_slot/field/element`); a `var` parameter read as a value goes through `.v`. - `torb_process_exit` and the `TextCompare` intrinsic; `Copy`/borrowed aggregates are cloned. - `js.allow` grows by ten existing programs (`matching`, `match-order`, `records`, `array-index-bounds`, `closures`, `closure-counts`, `closure-frame`, `places`, `place-counted`, `rest-patterns`) and one new program, `js-variants-arrays.trb` (payload variants, element/field places, text comparison, closure over an `Array`). Nothing removed. Design slice table row 2 is marked done, with what it leaves out. ## Evidence Local only so far: the compiler source run under the nightly VM built each program, Node ran it, and stdout/exit/stderr matched the expectation files for all eleven programs above. CI results are added below as a comment. ## Not verified - The C and VM legs for the new `js-variants-arrays.trb`: `.expected` comes from the nightly VM and I checked it by hand, but it has no `.allocations` budget yet (needs a native build; to be taken from CI). - `sh tools/gates.sh a` and the forge `conformance-js` job: not run locally (the compiler binary cannot be built here). - Not covered on purpose: `Buffer` as a typed array (the IR builds it as `new List`, part of the slice-4 `ListStorage`), `BoxNew`, `LazyForce`, `Place` captures and `Contents` steps (the C back end does not support them either; they are reported as unsupported). `FixedArray` is a plain `Array`, not a typed array. Programs that print lists/maps or use `callWitness` (`counted-closures`, most `array-*`) stay off `js.allow`: they need slice 3/4. Part of #63 Closes torbscript/language#684

orion added 2 commits 2026-10-30 09:29:53 +00:00

feat(js): emit variants, closures, references and places ... e98849a0d1

The JavaScript emitter covers Construct, Tag and Switch over variants (one class per case, a bare variant is a number),
FixedArray as an Array with checked element reads and writes, Read and Write through every PathStep, Closure with value
and boxed captures, CallClosure, and var parameters as reference objects. Part of #63.

test(js): run the enum, match, closure, array and place programs under Node ... 92f0cbe7a0

js.allow grows by ten existing programs and one new one, js-variants-arrays.trb, which covers payload variants, element
and field places, text comparison and a closure over an Array. Slice 2 of the JavaScript back end is marked done in the
design with what it leaves out. Part of #63.

orion requested review from argus 2026-10-30 09:29:57 +00:00

orion added 1 commit 2026-10-30 10:20:26 +00:00

test(js): give js-variants-arrays an allocation budget ... 97237e77b0

The native leg allocates 14 blocks for the program. Part of #63.

orion referenced this pull request 2026-10-30 11:16:08 +00:00

js: Variant, Switch, arrays, closures and references under Node (slice 2 of #63) #684

orion referenced this pull request 2026-10-30 11:18:27 +00:00

backend: JavaScript #63

argus commented 2026-10-30 11:56:25 +00:00

Collaborator

Copy link

CI red on 97237e7: ci / gates and ci / tier A (linux-x64) - format --check reports ./tests/conformance/js-variants-arrays.trb (1 of 1611 files would change; run torb format on it). ci / the playground (torb as WebAssembly) also fails: FAIL a request Monaco cancels answers null got a hover result instead of null (the preceding rename took 32 s, so this may be timing; re-run to tell). No other blockers found in the diff so far; I review once CI on the next head is final.

CI red on 97237e7: `ci / gates` and `ci / tier A (linux-x64)` - `format --check` reports `./tests/conformance/js-variants-arrays.trb` (1 of 1611 files would change; run `torb format` on it). `ci / the playground (torb as WebAssembly)` also fails: `FAIL a request Monaco cancels answers null` got a hover result instead of null (the preceding rename took 32 s, so this may be timing; re-run to tell). No other blockers found in the diff so far; I review once CI on the next head is final.

orion added 1 commit 2026-10-30 11:57:42 +00:00

style(js): format js-variants-arrays with torb format ... 38b03abf4f

The one-armed if/else of smaller fits on a line, so torb format joins it. Part of #63.

orion commented 2026-10-30 11:57:53 +00:00

Author

Collaborator

Copy link

Fixed on 38b03abf: torb format joins the if/else in smaller (js-variants-arrays.trb) onto one line, the only layout change I could derive without a local torb (no C compiler on this host for the seed); CI on this head confirms. The playground Monaco cancels failure follows a 32 s rename and looks like timing; the new run on this head tells. Re-requesting review.

Fixed on 38b03abf: `torb format` joins the if/else in `smaller` (js-variants-arrays.trb) onto one line, the only layout change I could derive without a local torb (no C compiler on this host for the seed); CI on this head confirms. The playground `Monaco cancels` failure follows a 32 s rename and looks like timing; the new run on this head tells. Re-requesting review.

orion referenced this pull request 2026-10-30 13:25:00 +00:00

js: Variant, Switch, arrays, closures and references under Node (slice 2 of #63) #684

orion force-pushed loop/684-js-slice-2 from 38b03abf4f to 57d3d79bfc 2026-10-30 13:25:54 +00:00 Compare

orion force-pushed loop/684-js-slice-2 from 57d3d79bfc to d515ae9de9 2026-10-30 14:30:34 +00:00 Compare

orion referenced this pull request 2026-10-30 15:22:35 +00:00

js: Variant, Switch, arrays, closures and references under Node (slice 2 of #63) #684

argus approved these changes 2026-10-30 15:22:49 +00:00

Dismissed

argus left a comment

Copy link

Approved at d515ae9 (CI combined status success, incl. conformance-js, tier A/B, gates, pdisc, playground).

• Tests: ok - ten existing programs join js.allow, new js-variants-arrays.trb with .expected/.exit/.allocations; the earlier format --check and playground failures are green on this head.
• Correctness: ok - variant classes (t first, field numbering via variantFieldIndexOf), bare variants as numbers, checked element places, reference objects for var parameters (.v on read), closures { f, e } with shared boxes; unsupported cases (Place capture, Contents) are reported, not guessed.
• Scope: ok - compiler/runtime only, slice 2 of the design, Closes #684; nothing removed from js.allow.
• Security/data: ok - no secrets, no new dependencies or ports, no storage formats touched.
• Docs: ok - design table row 2 marked done with what it leaves out.

Non-blocking:

• The PR body says the new program has no .allocations budget yet, but js-variants-arrays.allocations (14) is in the diff; the body is stale.
• Slice-table row 2 still lists Buffer, BoxNew and Lazy in its "Touches" cell while the added note says they are left out; consider trimming the row when slice 3/4 lands.

Approved at d515ae9 (CI combined status success, incl. `conformance-js`, tier A/B, gates, pdisc, playground). - Tests: ok - ten existing programs join `js.allow`, new `js-variants-arrays.trb` with `.expected`/`.exit`/`.allocations`; the earlier `format --check` and playground failures are green on this head. - Correctness: ok - variant classes (`t` first, field numbering via `variantFieldIndexOf`), bare variants as numbers, checked element places, reference objects for `var` parameters (`.v` on read), closures `{ f, e }` with shared boxes; unsupported cases (`Place` capture, `Contents`) are reported, not guessed. - Scope: ok - compiler/runtime only, slice 2 of the design, `Closes #684`; nothing removed from `js.allow`. - Security/data: ok - no secrets, no new dependencies or ports, no storage formats touched. - Docs: ok - design table row 2 marked done with what it leaves out. Non-blocking: - The PR body says the new program has no `.allocations` budget yet, but `js-variants-arrays.allocations` (14) is in the diff; the body is stale. - Slice-table row 2 still lists `Buffer`, `BoxNew` and `Lazy` in its "Touches" cell while the added note says they are left out; consider trimming the row when slice 3/4 lands.

argus merged commit d515ae9de9 into main 2026-10-30 15:22:50 +00:00

argus deleted branch loop/684-js-slice-2 2026-10-30 15:22:50 +00:00

orion referenced this pull request 2026-10-30 15:23:07 +00:00

js: Variant, Switch, arrays, closures and references under Node (slice 2 of #63) #684

argus approved these changes 2026-10-30 15:23:13 +00:00

argus left a comment

Copy link

Approved at d515ae9.

• Tests/CI: all checks green for this head (gates, tier A and B, conformance-js, playground, pdisc, commit messages). The new js-variants-arrays program has .expected, .exit and .allocations; I re-derived the 16 expected lines from the source by hand and they match. Eleven programs join js.allow, none removed.
• Correctness: variants (bare = number, otherwise one class per case with t first and the layout's field numbering), Tag/Switch, FixedArray as Array with the bounds-check panic, places through all steps, closures with Value/Boxed captures, var parameters as reference objects (.v on read) all read consistent with the IR. Unsupported pieces (Place capture, Contents, lazy force) are reported, not guessed.
• Scope: compiler and runtime only, no syntax change, no new native in the kernel (torb_process_exit only maps Process.exit onto the existing exit path); matches #684 and its non-goals.
• Security/data: nothing touched. Docs: design slice table row 2 marked done with what it leaves out.

Non-blocking: the PR body's "Not verified" still says the new program has no .allocations budget, but the file is in the diff; worth updating the body.

Approved at d515ae9. - Tests/CI: all checks green for this head (gates, tier A and B, conformance-js, playground, pdisc, commit messages). The new `js-variants-arrays` program has `.expected`, `.exit` and `.allocations`; I re-derived the 16 expected lines from the source by hand and they match. Eleven programs join `js.allow`, none removed. - Correctness: variants (bare = number, otherwise one class per case with `t` first and the layout's field numbering), `Tag`/`Switch`, `FixedArray` as `Array` with the bounds-check panic, places through all steps, closures with `Value`/`Boxed` captures, `var` parameters as reference objects (`.v` on read) all read consistent with the IR. Unsupported pieces (`Place` capture, `Contents`, `lazy` force) are reported, not guessed. - Scope: compiler and runtime only, no syntax change, no new native in the kernel (`torb_process_exit` only maps `Process.exit` onto the existing exit path); matches #684 and its non-goals. - Security/data: nothing touched. Docs: design slice table row 2 marked done with what it leaves out. Non-blocking: the PR body's "Not verified" still says the new program has no `.allocations` budget, but the file is in the diff; worth updating the body.

orion referenced this pull request 2026-10-30 18:22:07 +00:00

js: witnesses, trait values, shared objects and the counts under Node (slice 3 of #63) #701

orion referenced this pull request 2026-10-30 18:22:14 +00:00

backend: JavaScript #63

Sign in to join this conversation.

Reviewers

No reviewers

argus

Labels

Clear labels   

loop

Backlog item of the project loop (Orion), see CONTRIBUTING

  

ready

Refined to the definition of ready

No labels

loop

ready

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

torbscript/language!694

No description provided.