backend: quoted expressions in the native back end (WEB F4)

orion commented

2026-10-05 21:07:21 +00:00

Collaborator

Context

Expression.value and Expression.captures are planned natives.

Source

WEB.md F4

Acceptance criteria

Both natives in the C back end
Typed form bindings work natively

Dependencies

None known.

Notes

Adds or changes a native, syntax or a name the compiler looks up: land it in two commits (teach both forms, refresh the seed, then migrate), as CLAUDE.md "Seed and breaking changes" says.

Moved from the planning lists of the docs; the design record stays the specification.

## Context `Expression.value` and `Expression.captures` are planned natives. ## Source - [WEB.md F4](https://git.torb.dev/torbscript/language/src/commit/393ef2608e4a60b8ff76d8b6b408951b822d887d/docs/design/WEB.md#L926) ## Acceptance criteria - [ ] Both natives in the C back end - [ ] Typed form bindings work natively ## Dependencies None known. ## Notes Adds or changes a native, syntax or a name the compiler looks up: land it in two commits (teach both forms, refresh the seed, then migrate), as `CLAUDE.md` "Seed and breaking changes" says. _Moved from the planning lists of the docs; the design record stays the specification._

orion added this to the Milestone 10: breadth of the standard library milestone

2026-10-05 21:07:21 +00:00

orion added the

loop

label

2026-10-05 21:07:21 +00:00

orion commented

2026-10-09 05:38:52 +00:00

Author

Collaborator

[loop:torbscript] intake: in scope (author orion, label loop), source WEB.md F4, not a duplicate. Adds natives, so it needs the two-commit rule (part 1, seed publish, part 2). No dependencies; ready buffer full: #4, #163, #164, #223 are ready. next: waiting (refine queued, ready buffer full) until 2026-10-10T04:00:00Z

orion commented

2026-10-26 01:32:48 +00:00

Author

Collaborator

[loop:torbscript] claim refine by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z

orion commented

2026-10-26 01:34:17 +00:00

Author

Collaborator

[loop:torbscript] refine: the issue text is stale; only Expression.captures() and the typed-binding proof are open, next: ready

Verified on main (c72cabec)

Expression.value() is done: both back ends build an Expression<Value> as a record (tree, source, location, lazy value cell), compiler/src/ir/lower/quoted.trb, value() is an intrinsic in ir/lower/call.trb (no C native), held by tests/conformance/quotation-values.trb. WEB.md problem 5 is marked closed for that part; only the plannedRuntimeOf("Expression.value"/"Expression.captures", ...) rows in compiler/src/backend/c/natives.trb still read "planned".
Open: captures() (quoted.trb header: "is not built"; docs/language/functions/quoted-expressions.md banner). The checker already requires every capture to be Encode (semantics/checker/quote.trb, encodedCaptures), so each capture can become an EncodedValue through EncodedValue.of (std/encoding/src/values.trb).
Dependencies: none open. F2, F3, F5, F6 of WEB.md are done; std/web does not exist yet, so "typed form bindings" is proven with a user-defined fn input(bind: Expression<String>) in a conformance test.

decided by the loop: captures() stays an intrinsic (like value()), not a new C native: no runtime/ change, no new driver flag, no syntax, std/expression declarations unchanged. The two-commit rule therefore does not apply (single commit feat(backend): build Expression.captures natively). If the worker finds it must change a std/ declaration the compiler's own build reads, it stops and comments instead of going on. assert keeps its own special lowering (ir/lower/quote.trb) and is not touched. The captures list is a second lazy cell next to value, so a program that never calls captures() pays no run time; the extra Encode instances per captured type are measured in forge CI (binary size of the compiler/test targets) and accepted if the growth is small.

Acceptance criteria (replace the issue's)

expression.captures() answers List<EncodedValue> in Captured.index order, natively and in the VM (torb run, torb test), for scalars, String, a record and a list capture; a quotation without captures answers [].
A user-defined fn input(bind: Expression<String>) called as input bind: form.email builds natively and answers nameOf(bind), bind.value() and bind.captures() (the typed form binding of WEB.md 3.5).
captures() is lazy: not evaluated unless asked; capture values are the ones at the quotation site, read once.
plannedRuntimeOf("Expression.value"/"Expression.captures") rows gone from natives.trb (value is an intrinsic); docs banner and WEB.md problem 5 / F4 updated, skills/ regenerated.

Steps (one PR, each step one worker session)

compiler/src/ir/lower/quoted.trb (+ ir/instantiate.trb layout of Expression, treeBuilderOf field count): add the hidden fifth field, a lazy cell of List<EncodedValue> built from quotation.captures via a monomorphised EncodedValue.of<T> per capture type (see lowerLazyArgument, listOfValues); fix the fields.length() != 4 check.
compiler/src/ir/lower/call.trb: owner == "Expression" && name == "captures" next to value, forcing the new cell (lowerQuotedValue pattern); drop the two planned rows in backend/c/natives.trb; check the VM path through the IR (no VM change expected).
Tests and docs: new tests/conformance/quotation-captures.trb (+ .expected, .exit, .allocations), a typed-binding case from criterion 2, extend compiler/tests/quotations.test.trb only if lowering text is asserted; update docs/language/functions/quoted-expressions.md (remove the captures part of the banner, keep the other unsupported nodes listed: map literal, range, implicit-self field, non-constant top-level const), docs/standard-library/expression.md, WEB.md problem 5 and F4 row, ENCODING.md slice 5 row; regenerate skills/.
Out of scope (no issue yet, just in time): the four unsupported node kinds above.
Gates run in forge CI only; locally torb format --check on touched files and one conformance test.

[loop:torbscript] refine: the issue text is stale; only `Expression.captures()` and the typed-binding proof are open, next: ready **Verified on main (c72cabec)** - `Expression.value()` is done: both back ends build an `Expression<Value>` as a record (tree, source, location, `lazy` value cell), `compiler/src/ir/lower/quoted.trb`, `value()` is an intrinsic in `ir/lower/call.trb` (no C native), held by `tests/conformance/quotation-values.trb`. WEB.md problem 5 is marked closed for that part; only the `plannedRuntimeOf("Expression.value"/"Expression.captures", ...)` rows in `compiler/src/backend/c/natives.trb` still read "planned". - Open: `captures()` (`quoted.trb` header: "is not built"; `docs/language/functions/quoted-expressions.md` banner). The checker already requires every capture to be `Encode` (`semantics/checker/quote.trb`, `encodedCaptures`), so each capture can become an `EncodedValue` through `EncodedValue.of` (`std/encoding/src/values.trb`). - Dependencies: none open. F2, F3, F5, F6 of WEB.md are done; `std/web` does not exist yet, so "typed form bindings" is proven with a user-defined `fn input(bind: Expression<String>)` in a conformance test. **decided by the loop:** `captures()` stays an intrinsic (like `value()`), not a new C native: no `runtime/` change, no new driver flag, no syntax, `std/expression` declarations unchanged. The two-commit rule therefore does not apply (single commit `feat(backend): build Expression.captures natively`). If the worker finds it must change a `std/` declaration the compiler's own build reads, it stops and comments instead of going on. `assert` keeps its own special lowering (`ir/lower/quote.trb`) and is not touched. The captures list is a second `lazy` cell next to `value`, so a program that never calls `captures()` pays no run time; the extra `Encode` instances per captured type are measured in forge CI (binary size of the compiler/test targets) and accepted if the growth is small. **Acceptance criteria (replace the issue's)** 1. `expression.captures()` answers `List<EncodedValue>` in `Captured.index` order, natively and in the VM (`torb run`, `torb test`), for scalars, `String`, a record and a list capture; a quotation without captures answers `[]`. 2. A user-defined `fn input(bind: Expression<String>)` called as `input bind: form.email` builds natively and answers `nameOf(bind)`, `bind.value()` and `bind.captures()` (the typed form binding of WEB.md 3.5). 3. `captures()` is lazy: not evaluated unless asked; capture values are the ones at the quotation site, read once. 4. `plannedRuntimeOf("Expression.value"/"Expression.captures")` rows gone from `natives.trb` (value is an intrinsic); docs banner and WEB.md problem 5 / F4 updated, `skills/` regenerated. **Steps (one PR, each step one worker session)** 1. `compiler/src/ir/lower/quoted.trb` (+ `ir/instantiate.trb` layout of `Expression`, `treeBuilderOf` field count): add the hidden fifth field, a `lazy` cell of `List<EncodedValue>` built from `quotation.captures` via a monomorphised `EncodedValue.of<T>` per capture type (see `lowerLazyArgument`, `listOfValues`); fix the `fields.length() != 4` check. 2. `compiler/src/ir/lower/call.trb`: `owner == "Expression" && name == "captures"` next to `value`, forcing the new cell (`lowerQuotedValue` pattern); drop the two planned rows in `backend/c/natives.trb`; check the VM path through the IR (no VM change expected). 3. Tests and docs: new `tests/conformance/quotation-captures.trb` (+ `.expected`, `.exit`, `.allocations`), a typed-binding case from criterion 2, extend `compiler/tests/quotations.test.trb` only if lowering text is asserted; update `docs/language/functions/quoted-expressions.md` (remove the captures part of the banner, keep the other unsupported nodes listed: map literal, range, implicit-`self` field, non-constant top-level `const`), `docs/standard-library/expression.md`, WEB.md problem 5 and F4 row, ENCODING.md slice 5 row; regenerate `skills/`. Out of scope (no issue yet, just in time): the four unsupported node kinds above. Gates run in forge CI only; locally `torb format --check` on touched files and one conformance test.

orion added the

ready

label

2026-10-26 01:34:19 +00:00

orion commented

2026-10-26 01:42:50 +00:00

Author

Collaborator

[loop:torbscript] claim implement by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z

orion commented

2026-10-26 01:44:42 +00:00

Author

Collaborator

[loop:torbscript] implement: checkpoint, design explored, no code written yet (context guard), no PR yet, next: implement (re-dispatch; the claim stays live)

Findings for the next worker (all on main c72cabec, compiler/src/ir/):

Layout: ir/instantiate.trb ~l.431 appends the hidden value cell field to Expression<Value>; add a second field captures with cellType program, checker, instantiation, <List<EncodedValue> TypeId>, <its IrType>. Then fix the fields.length() != 4 checks (lower/quoted.trb treeBuilderOf, lowerQuotedValue) to 5, and Construct fields in lowerQuotation (add the second cell).
Thunk for the cell: lower/closure.trb lowerLazyArgument takes an AST Expression (ClosureSource.Thunk), so a captures thunk needs a third ClosureSource case (e.g. Captures(quotation)) handled in lowerClosureBody (l.1276), loweredClosureBody (taskValue match, l.~1157) and the parameter-binding match (l.~1350). Split lowerLazyArgument so the cell building (cellTypeFor, emptyOptionSlot, Construct) is shared; call closureValueOf with at = the quotation span (plans come from lowering.quotedCaptures.get(at), already set in lowerQuotation, same order as Captured.index).
Body of that thunk: after bindCaptures, for each quotation.captures binding read its slot (lowering.slots.get(declared), or bindingReferenceOf for boxed) and emit a call to EncodedValue.of<T>: lookupStaticMember(checker, encodedValueType, "of") -> .Method(_, symbol); mapping = memberMappingOf(...) (ir/witness.trb) plus mapping.set(signature.generics.last(), captureType) for Value; instanceFor lowering, symbol, mapping (ir/instances.trb); emitCall with operands [value, bool false] (the default privateFields must be passed explicitly; StaticContents.Literal(LiteralValue.Boolean(false)) via constantSlot). Pattern to copy: emitWithCapacity in lower/collection.trb l.285. Collect the results with listOfValues(lowering, builder, slots, encodedValueTypeId, listIrType, at) and return it. EncodedValue TypeId via preludeType(lowering, "EncodedValue", []) (report unsupported if absent).
captures() call: lower/call.trb, next to the Expression.value intrinsic; add lowerQuotedCaptures in quoted.trb mirroring lowerQuotedValue (field 4 instead of 3, lowerForcedRead).
Then: drop the planned Expression.value/Expression.captures rows in backend/c/natives.trb; tests tests/conformance/quotation-captures.trb (+ .expected/.exit/.allocations) and the input bind: form.email case; docs banner, docs/standard-library/expression.md, WEB.md problem 5/F4, ENCODING.md slice 5, regenerate skills/. Single commit feat(backend): build Expression.captures natively. Gates in forge CI only.
Worktree removed (nothing was written); recreate with ws wt new torbscript/language 106 expression-captures --branch loop/106-expression-captures.

[loop:torbscript] implement: checkpoint, design explored, no code written yet (context guard), no PR yet, next: implement (re-dispatch; the claim stays live) Findings for the next worker (all on main c72cabec, `compiler/src/ir/`): - Layout: `ir/instantiate.trb` ~l.431 appends the hidden `value` cell field to `Expression<Value>`; add a second field `captures` with `cellType program, checker, instantiation, <List<EncodedValue> TypeId>, <its IrType>`. Then fix the `fields.length() != 4` checks (`lower/quoted.trb` `treeBuilderOf`, `lowerQuotedValue`) to 5, and `Construct` fields in `lowerQuotation` (add the second cell). - Thunk for the cell: `lower/closure.trb` `lowerLazyArgument` takes an AST `Expression` (`ClosureSource.Thunk`), so a captures thunk needs a third `ClosureSource` case (e.g. `Captures(quotation)`) handled in `lowerClosureBody` (l.1276), `loweredClosureBody` (`taskValue` match, l.~1157) and the parameter-binding match (l.~1350). Split `lowerLazyArgument` so the cell building (cellTypeFor, emptyOptionSlot, Construct) is shared; call `closureValueOf` with `at` = the quotation span (plans come from `lowering.quotedCaptures.get(at)`, already set in `lowerQuotation`, same order as `Captured.index`). - Body of that thunk: after `bindCaptures`, for each `quotation.captures` binding read its slot (`lowering.slots.get(declared)`, or `bindingReferenceOf` for boxed) and emit a call to `EncodedValue.of<T>`: `lookupStaticMember(checker, encodedValueType, "of")` -> `.Method(_, symbol)`; `mapping = memberMappingOf(...)` (ir/witness.trb) plus `mapping.set(signature.generics.last(), captureType)` for `Value`; `instanceFor lowering, symbol, mapping` (ir/instances.trb); `emitCall` with operands [value, bool false] (the default `privateFields` must be passed explicitly; `StaticContents.Literal(LiteralValue.Boolean(false))` via `constantSlot`). Pattern to copy: `emitWithCapacity` in `lower/collection.trb` l.285. Collect the results with `listOfValues(lowering, builder, slots, encodedValueTypeId, listIrType, at)` and return it. `EncodedValue` TypeId via `preludeType(lowering, "EncodedValue", [])` (report unsupported if absent). - `captures()` call: `lower/call.trb`, next to the `Expression.value` intrinsic; add `lowerQuotedCaptures` in `quoted.trb` mirroring `lowerQuotedValue` (field 4 instead of 3, `lowerForcedRead`). - Then: drop the planned `Expression.value`/`Expression.captures` rows in `backend/c/natives.trb`; tests `tests/conformance/quotation-captures.trb` (+ .expected/.exit/.allocations) and the `input bind: form.email` case; docs banner, `docs/standard-library/expression.md`, WEB.md problem 5/F4, ENCODING.md slice 5, regenerate `skills/`. Single commit `feat(backend): build Expression.captures natively`. Gates in forge CI only. Worktree removed (nothing was written); recreate with `ws wt new torbscript/language 106 expression-captures --branch loop/106-expression-captures`.

orion commented

2026-10-26 01:45:50 +00:00

Author

Collaborator

[loop:torbscript] implement: re-dispatched from checkpoint #6732 (continuation 2 of 3, light bootstrap to stay under the context guard), claim stays live, next: implement (worker running)

orion referenced this issue from a commit

2026-10-26 01:48:54 +00:00

feat(backend): build Expression.captures natively

orion commented

2026-10-26 01:49:15 +00:00

Author

Collaborator

[loop:torbscript] implement: checkpoint 2, branch loop/106-expression-captures pushed (WIP commit 53ba5a4d, amend it, keep one commit), no PR yet, next: implement (continuation 3 of 3; the claim stays live)

Done (in 53ba5a4d)

compiler/src/ir/instantiate.trb: buildNominalLayout appends a 5th field captures (cell of List<EncodedValue>) after value; new public fn capturesTypeOf(var checker: Checker): TypeId? = TypeForm.Traits([nominal(List, [nominal(EncodedValue, [])])]) (the same interned TypeId std/expression declares as the result of captures()). Not compiled yet: check checker.types.intern / checker.wellKnown.symbolNamed are callable from there (symbolNamed is fn on WellKnown l.137).

Left, exact recipe (all verified against main, no more exploring needed)

ir/lower/closure.trb: type ClosureSource (l.261) add case Captures(quotation: Quotation, at: Span) (+ use Quotation from "../../semantics/checker/context"). Matches: loweredClosureBody taskValue (l.~1157) .Thunk(_) | .Captures(_, _) => None; bindClosureParameters (l.~1350) same, return; lowerClosureBody (l.1276) .Captures(quotation, at) => call capturesListOf(lowering, builder, quotation, result, at) (new, in quoted.trb), return unless supported, finishBlock Terminator.Return(borrowed(slot)). Split lowerLazyArgument (l.307) into a private lazyCellOf(lowering, builder, source: ClosureSource, item: TypeId, at: Span): Slot? (all of its current body, value.span -> at) and add public fn lowerCapturesCell(lowering, builder, quotation, item, at) = lazyCellOf(..., ClosureSource.Captures(quotation, at), ...). capturePlansOf already finds the plans through lowering.quotedCaptures.get(at) (set in lowerQuotation, not removed).
ir/lower/quoted.trb: new public fn capturesListOf(var lowering, var builder, quotation: Quotation, annotation: IrTypeId, at: Span): Slot?: encoded = preludeTypeOrReport(lowering, "EncodedValue", at)?; lookupStaticMember(lowering.checker, encoded, "of") (semantics/checker/member) -> .Method(_, symbol); parameter = lowering.checker.signature(signatureOf(lowering.checker, symbol)).generics.last(); privateFields = constantSlot(lowering, builder, None, StaticContents.Literal(LiteralValue.Boolean(false)), booleanTypeOf(lowering)); per capture in quotation.captures: declared = lowering.checker.binding(capture.binding).at; slot = lowering.slots.get(declared), else bindingReferenceOf(lowering, builder, declared) (closure.trb l.249) + Read into a temporary; mapping = memberMappingOf(lowering, symbol, encoded, encoded) (../witness), mapping.set parameter, substituted(lowering, capture.type); function = instanceFor(lowering, symbol, mapping)? (../instances); emitCall lowering, builder, None, function, [slot, privateFields], irTypeOfClosed(lowering, encoded), at (./call, l.1060). End: listOfValues lowering, builder, values, encoded, annotation, at.
In lowerQuotation after the value cell: item = capturesTypeOf(lowering.checker) (report unsupported if None), lowerCapturesCell lowering, builder, quotation, item, at, add borrowed(capturesCell) as 5th Construct field. treeBuilderOf (l.142) and lowerQuotedValue (l.803): != 4 -> != 5 (+ message "... value and captures"). Add lowerQuotedCaptures = lowerQuotedValue with field 4 (factor both into one forcedFieldOf(..., index, what)). Update the module doc comment (l.1-17: captures() is now built, second lazy cell).
ir/lower/call.trb l.373: next to value, if owner == "Expression" && name == "captures" && operands.length() == 1 { return lowerQuotedCaptures ... } (+ import l.57).
Finding: do NOT just delete the planned rows: value/captures are native fn in std/expression, and compiler/tests/natives-audit.test.trb requires a manifest entry per std native. Replace the two plannedRuntimeOf(...) rows in backend/c/natives.trb l.1988-1989 with loweredOf("Expression.value") / loweredOf("Expression.captures") (target .Lowered, like spawn); in docs/design/NATIVES.md l.850-851 set Target Lowered (verdict keep, group compiler unchanged, so the Counts section stays). Also docs/BACKEND.md l.3562 says they are .Planned("5.11"): fix.
Tests/docs per refine #6713 step 3: tests/conformance/quotation-captures.trb (+ .expected/.exit/.allocations, copy the shape of quotation-values.*), fn input(bind: Expression<String>) case; docs/language/functions/quoted-expressions.md banner, docs/standard-library/expression.md, WEB.md problem 5/F4, ENCODING.md l.937 slice 5 row; regenerate skills/ (a nightly torb exists at /tmp/torb110/x/torb-nightly-20261025-linux-x64/bin/torb: torb docs skill docs skills, torb format --check <files>).
Amend the single commit feat(backend): build Expression.captures natively, force-push the branch, open the PR (Closes #106), one blocking CI wait.

Worktree: /home/node/workspaces/orion/code/wt/torbscript-language-106-expression-captures (kept). Worker usage: ~90k prompt tokens (context guard at 80k).

[loop:torbscript] implement: checkpoint 2, branch `loop/106-expression-captures` pushed (WIP commit 53ba5a4d, amend it, keep one commit), no PR yet, next: implement (continuation 3 of 3; the claim stays live) **Done (in 53ba5a4d)** - `compiler/src/ir/instantiate.trb`: `buildNominalLayout` appends a 5th field `captures` (cell of `List<EncodedValue>`) after `value`; new `public fn capturesTypeOf(var checker: Checker): TypeId?` = `TypeForm.Traits([nominal(List, [nominal(EncodedValue, [])])])` (the same interned TypeId `std/expression` declares as the result of `captures()`). Not compiled yet: check `checker.types.intern` / `checker.wellKnown.symbolNamed` are callable from there (`symbolNamed` is `fn` on WellKnown l.137). **Left, exact recipe (all verified against main, no more exploring needed)** 1. `ir/lower/closure.trb`: `type ClosureSource` (l.261) add `case Captures(quotation: Quotation, at: Span)` (+ `use Quotation from "../../semantics/checker/context"`). Matches: `loweredClosureBody` taskValue (l.~1157) `.Thunk(_) | .Captures(_, _) => None`; `bindClosureParameters` (l.~1350) same, `return`; `lowerClosureBody` (l.1276) `.Captures(quotation, at) =>` call `capturesListOf(lowering, builder, quotation, result, at)` (new, in quoted.trb), return unless supported, `finishBlock Terminator.Return(borrowed(slot))`. Split `lowerLazyArgument` (l.307) into a private `lazyCellOf(lowering, builder, source: ClosureSource, item: TypeId, at: Span): Slot?` (all of its current body, `value.span` -> `at`) and add `public fn lowerCapturesCell(lowering, builder, quotation, item, at) = lazyCellOf(..., ClosureSource.Captures(quotation, at), ...)`. `capturePlansOf` already finds the plans through `lowering.quotedCaptures.get(at)` (set in `lowerQuotation`, not removed). 2. `ir/lower/quoted.trb`: new `public fn capturesListOf(var lowering, var builder, quotation: Quotation, annotation: IrTypeId, at: Span): Slot?`: `encoded = preludeTypeOrReport(lowering, "EncodedValue", at)?`; `lookupStaticMember(lowering.checker, encoded, "of")` (`semantics/checker/member`) -> `.Method(_, symbol)`; `parameter = lowering.checker.signature(signatureOf(lowering.checker, symbol)).generics.last()`; `privateFields = constantSlot(lowering, builder, None, StaticContents.Literal(LiteralValue.Boolean(false)), booleanTypeOf(lowering))`; per `capture in quotation.captures`: `declared = lowering.checker.binding(capture.binding).at`; slot = `lowering.slots.get(declared)`, else `bindingReferenceOf(lowering, builder, declared)` (closure.trb l.249) + `Read` into a temporary; `mapping = memberMappingOf(lowering, symbol, encoded, encoded)` (`../witness`), `mapping.set parameter, substituted(lowering, capture.type)`; `function = instanceFor(lowering, symbol, mapping)?` (`../instances`); `emitCall lowering, builder, None, function, [slot, privateFields], irTypeOfClosed(lowering, encoded), at` (`./call`, l.1060). End: `listOfValues lowering, builder, values, encoded, annotation, at`. In `lowerQuotation` after the value cell: `item = capturesTypeOf(lowering.checker)` (report unsupported if None), `lowerCapturesCell lowering, builder, quotation, item, at`, add `borrowed(capturesCell)` as 5th Construct field. `treeBuilderOf` (l.142) and `lowerQuotedValue` (l.803): `!= 4` -> `!= 5` (+ message "... value and captures"). Add `lowerQuotedCaptures` = `lowerQuotedValue` with field 4 (factor both into one `forcedFieldOf(..., index, what)`). Update the module doc comment (l.1-17: `captures()` is now built, second lazy cell). 3. `ir/lower/call.trb` l.373: next to `value`, `if owner == "Expression" && name == "captures" && operands.length() == 1 { return lowerQuotedCaptures ... }` (+ import l.57). 4. **Finding:** do NOT just delete the planned rows: `value`/`captures` are `native fn` in `std/expression`, and `compiler/tests/natives-audit.test.trb` requires a manifest entry per std native. Replace the two `plannedRuntimeOf(...)` rows in `backend/c/natives.trb` l.1988-1989 with `loweredOf("Expression.value")` / `loweredOf("Expression.captures")` (target `.Lowered`, like `spawn`); in `docs/design/NATIVES.md` l.850-851 set Target `Lowered` (verdict `keep`, group `compiler` unchanged, so the Counts section stays). Also `docs/BACKEND.md` l.3562 says they are `.Planned("5.11")`: fix. 5. Tests/docs per refine #6713 step 3: `tests/conformance/quotation-captures.trb` (+ `.expected/.exit/.allocations`, copy the shape of `quotation-values.*`), `fn input(bind: Expression<String>)` case; `docs/language/functions/quoted-expressions.md` banner, `docs/standard-library/expression.md`, WEB.md problem 5/F4, ENCODING.md l.937 slice 5 row; regenerate `skills/` (a nightly torb exists at `/tmp/torb110/x/torb-nightly-20261025-linux-x64/bin/torb`: `torb docs skill docs skills`, `torb format --check <files>`). 6. Amend the single commit `feat(backend): build Expression.captures natively`, force-push the branch, open the PR (`Closes #106`), one blocking CI wait. Worktree: `/home/node/workspaces/orion/code/wt/torbscript-language-106-expression-captures` (kept). Worker usage: ~90k prompt tokens (context guard at 80k).

orion referenced this issue from a commit

2026-10-26 01:51:23 +00:00

feat(backend): build Expression.captures natively

orion referenced this issue from a commit

2026-10-26 01:51:50 +00:00

feat(backend): build Expression.captures natively

orion referenced this issue from a pull request that will close it,

2026-10-26 01:52:21 +00:00

feat(backend): build Expression.captures natively #495

orion referenced this issue from a commit

2026-10-26 01:52:22 +00:00

feat(backend): build Expression.captures natively

orion commented

2026-10-26 02:02:39 +00:00

Author

Collaborator

[loop:torbscript] implement: Expression.captures built natively (hidden 5th captures field and lazy cell, capturesListOf, intrinsic in call.trb, loweredOf manifest rows, NATIVES/BACKEND/ENCODING/WEB/quoted-expressions docs, skills regenerated, conformance quotation-captures), PR #495 (single commit 661fb742, Argus review requested), next: waiting (forge CI run 1826 pending after 10 min)

Remainder, to be fixed in the same commit once CI reports

Nothing was compiled or run locally: the first CI run may show compile errors in ir/lower/quoted.trb / closure.trb (the nightly torb could not parse them, so torb format --check is unchecked for both), fix on the same branch.
tests/conformance/quotation-captures.allocations is missing (CI prints the count; unbudgeted only with a reason) and quotation-values.allocations (302) will rise by the cells of its quotations: set both from the CI counts, with a commit message line saying why.
quotation-captures.expected was written by hand: the order 100, 18 for base + minimum + base and "Grace" for a String capture are assumptions; correct them from the CI diff.

Worktree code/wt/torbscript-language-106-expression-captures is pushed and kept for the fix round.

[loop:torbscript] implement: Expression.captures built natively (hidden 5th `captures` field and lazy cell, `capturesListOf`, intrinsic in call.trb, `loweredOf` manifest rows, NATIVES/BACKEND/ENCODING/WEB/quoted-expressions docs, skills regenerated, conformance `quotation-captures`), PR https://git.torb.dev/torbscript/language/pulls/495 (single commit 661fb742, Argus review requested), next: waiting (forge CI run 1826 pending after 10 min) **Remainder, to be fixed in the same commit once CI reports** - Nothing was compiled or run locally: the first CI run may show compile errors in `ir/lower/quoted.trb` / `closure.trb` (the nightly torb could not parse them, so `torb format --check` is unchecked for both), fix on the same branch. - `tests/conformance/quotation-captures.allocations` is missing (CI prints the count; `unbudgeted` only with a reason) and `quotation-values.allocations` (302) will rise by the cells of its quotations: set both from the CI counts, with a commit message line saying why. - `quotation-captures.expected` was written by hand: the order `100, 18` for `base + minimum + base` and `"Grace"` for a `String` capture are assumptions; correct them from the CI diff. Worktree `code/wt/torbscript-language-106-expression-captures` is pushed and kept for the fix round.

orion commented

2026-10-26 03:44:41 +00:00

Author

Collaborator

[loop:torbscript] claim fix by session:loop-torbscript-20261026 until 2026-10-26T08:00:00Z (PR 495 bootstrap red)

orion referenced this issue from a commit

2026-10-26 03:46:32 +00:00

feat(backend): build Expression.captures natively

orion commented

2026-10-26 03:57:42 +00:00

Author

Collaborator

[loop:torbscript] fix: the bootstrap compile errors are fixed in the single commit (now bbeba43b, force-pushed): closure.trb called the enum case ClosureSource.Thunk value without parentheses, quoted.trb had match capturesTypeOf lowering.checker { (needs parentheses) and used type as a binding name (keyword, now capturedType); the nightly torb parses both files and torb format --check is clean; check-commits.sh clean; Argus review re-requested. next: waiting (forge CI run 1841 still queued after 11 min, no job started)

Remainder, to be fixed in the same commit once CI reports

The first run that actually compiles may show further type errors in ir/lower/quoted.trb / closure.trb / instantiate.trb (never compiled before; the nightly torb is too old to check them): fix on the same branch.
tests/conformance/quotation-captures.allocations is still missing (CI prints the count; unbudgeted only with a reason) and quotation-values.allocations (302) will rise: set both from the CI counts, with a commit message line saying why.
quotation-captures.expected was written by hand (order 100, 18 for base + minimum + base, "Grace" for the String capture): correct it from the CI diff.

Worktree code/wt/torbscript-language-106-expression-captures is removed; recreate it from loop/106-expression-captures.

[loop:torbscript] fix: the bootstrap compile errors are fixed in the single commit (now bbeba43b, force-pushed): `closure.trb` called the enum case `ClosureSource.Thunk value` without parentheses, `quoted.trb` had `match capturesTypeOf lowering.checker {` (needs parentheses) and used `type` as a binding name (keyword, now `capturedType`); the nightly torb parses both files and `torb format --check` is clean; `check-commits.sh` clean; Argus review re-requested. next: waiting (forge CI run 1841 still queued after 11 min, no job started) **Remainder, to be fixed in the same commit once CI reports** - The first run that actually compiles may show further type errors in `ir/lower/quoted.trb` / `closure.trb` / `instantiate.trb` (never compiled before; the nightly torb is too old to check them): fix on the same branch. - `tests/conformance/quotation-captures.allocations` is still missing (CI prints the count; `unbudgeted` only with a reason) and `quotation-values.allocations` (302) will rise: set both from the CI counts, with a commit message line saying why. - `quotation-captures.expected` was written by hand (order `100, 18` for `base + minimum + base`, `"Grace"` for the `String` capture): correct it from the CI diff. Worktree `code/wt/torbscript-language-106-expression-captures` is removed; recreate it from `loop/106-expression-captures`.

orion commented

2026-10-26 07:17:10 +00:00

Author

Collaborator

[loop:torbscript] claim fix by session:loop-torbscript-20261026-0724 until 2026-10-26T13:16:00Z (PR 495: not mergeable after main moved to 4f8e15cf, rebase and fix CI findings)

[loop:torbscript] claim fix by session:loop-torbscript-20261026-0724 until 2026-10-26T13:16:00Z (PR 495: not mergeable after main moved to 4f8e15cf, rebase and fix CI findings)

orion referenced this issue from a commit

2026-10-26 07:18:20 +00:00

feat(backend): build Expression.captures natively

orion commented

2026-10-26 08:45:14 +00:00

Author

Collaborator

[loop:torbscript] fix (checkpoint): rebased the single commit onto main 4f8e15cf (one conflict in compiler/src/ir/lower/call.trb, resolved by keeping both the Expression.captures case and main's nativeEntryOf call), new head 3a0c2f56, tools/check-commits.sh clean, force-pushed to loop/106-expression-captures. Forge CI of the new head: bootstrap (linux-x64) is green (the earlier compile errors are gone); tier B is red on three gates, all from this change: (1) tests/conformance/quotation-captures.trb:15 does not compile, "{nameOf bind} = {bind.value()}" reports Cannot infer Value of nameOf plus Expected } after the expression (command-call inside an interpolation; write {nameOf(bind)} and check whether nameOf needs a type argument), failing in the native build and in both VM lanes; (2) tests/conformance/quotation-values.trb allocates 330 blocks against its budget of 302 (+28), so either the captures cell is built eagerly where it should be lazy, or the budget needs sh tools/conformance.sh --update-budgets --filter quotation-values after checking the cause. gates, tier A and the other jobs were still pending. Argus review not re-requested yet (CI red). Queue wait was about 55 min before the first job started. Token usage of this worker: about 100k. next: fix round continues in worktree torbscript-language-106 (waiting for a continuation worker)

[loop:torbscript] fix (checkpoint): rebased the single commit onto main 4f8e15cf (one conflict in `compiler/src/ir/lower/call.trb`, resolved by keeping both the `Expression.captures` case and main's `nativeEntryOf` call), new head 3a0c2f56, `tools/check-commits.sh` clean, force-pushed to `loop/106-expression-captures`. Forge CI of the new head: `bootstrap (linux-x64)` is green (the earlier compile errors are gone); `tier B` is red on three gates, all from this change: (1) `tests/conformance/quotation-captures.trb:15` does not compile, `"{nameOf bind} = {bind.value()}"` reports `Cannot infer Value of nameOf` plus `Expected } after the expression` (command-call inside an interpolation; write `{nameOf(bind)}` and check whether `nameOf` needs a type argument), failing in the native build and in both VM lanes; (2) `tests/conformance/quotation-values.trb` allocates 330 blocks against its budget of 302 (+28), so either the captures cell is built eagerly where it should be lazy, or the budget needs `sh tools/conformance.sh --update-budgets --filter quotation-values` after checking the cause. `gates`, `tier A` and the other jobs were still pending. Argus review not re-requested yet (CI red). Queue wait was about 55 min before the first job started. Token usage of this worker: about 100k. next: fix round continues in worktree `torbscript-language-106` (waiting for a continuation worker)

orion referenced this issue from a commit

2026-10-26 08:47:32 +00:00

feat(backend): build Expression.captures natively

orion commented

2026-10-26 09:15:05 +00:00

Author

Collaborator

Checkpoint (Orion loop): head e3248733 on loop/106-expression-captures, PR 495. Fixed the two tier B failures: (1) quotation-captures.trb now writes {nameOf(bind)} (a command call is not valid inside an interpolation); (2) quotation-values allocated 330 vs 302 because every quotation got a memo cell (+ an environment when it captures) for captures() whether asked or not. captures() is now a plain thunk field with no cell: a capture-free quotation makes no block for it, a capturing one only its environment. Budget untouched; if the count still differs, CI names it and I adjust. CI is queued on the shared runners (gates/pdisc pending), tier A/B/bootstrap not started yet. Next: wait for full green, then re-request Argus.

Checkpoint (Orion loop): head e3248733 on loop/106-expression-captures, PR 495. Fixed the two tier B failures: (1) quotation-captures.trb now writes `{nameOf(bind)}` (a command call is not valid inside an interpolation); (2) quotation-values allocated 330 vs 302 because every quotation got a memo cell (+ an environment when it captures) for `captures()` whether asked or not. `captures()` is now a plain thunk field with no cell: a capture-free quotation makes no block for it, a capturing one only its environment. Budget untouched; if the count still differs, CI names it and I adjust. CI is queued on the shared runners (gates/pdisc pending), tier A/B/bootstrap not started yet. Next: wait for full green, then re-request Argus.

orion commented

2026-10-26 09:15:30 +00:00

Author

Collaborator

[loop:torbscript] release fix: the fix worker pushed e3248733 (quotation-captures interpolation fix, captures() thunk without a memo cell) and ended while forge CI of that head was still queued; remainder per the checkpoint above (adjust quotation-values.allocations if tier B reports a count above 302, then re-request Argus). next: waiting (CI on PR 495 head e3248733)

[loop:torbscript] release fix: the fix worker pushed e3248733 (quotation-captures interpolation fix, captures() thunk without a memo cell) and ended while forge CI of that head was still queued; remainder per the checkpoint above (adjust quotation-values.allocations if tier B reports a count above 302, then re-request Argus). next: waiting (CI on PR 495 head e3248733)

orion referenced this issue from a commit

2026-10-26 10:22:21 +00:00

test(conformance): set the allocation budgets of quotation-captures and quotation-values