feat(ir): lower native fn bodies through the Intrinsic path #487

Merged

argus merged 3 commits from loop/484-native-body-lowering into main 2026-10-26 06:29:00 +00:00

Conversation 3 Commits 3 Files changed 6 +258 -33

orion commented 2026-10-25 23:35:58 +00:00

Collaborator

Copy link

Part of #484 (slice 3a of the epic; slice 3 of docs/design/NATIVE-BODIES.md, section 8).

What this changes

• ir/intrinsic-name.trb: shared reading of intrinsic names, used by the checker (messages unchanged) and by lowering; the written .unchecked maps to checked: false.
• ir/lower/native-body.trb: a native fn body of exactly one intrinsic over self and the parameters in declared order becomes the NativeEntry of the call site (lowerNativeCall, lowerNativeOperator) and of the generated wrapper (generatedKindOf). A manifest row still wins; N9 is unchanged.

Not in this PR (epic checklist on #484)

NativeBlock lowering (needs operand resolution recorded by the checker first), multi-statement / const bodies, the missing-intrinsic finding, the record status line, the kernel test.

Follow-up on CI (b59107b)

Tier A failed on the two new lowering tests and format --check: declaredOperandNames counted self twice (the parser already puts the receiver back as the first parameter), and intrinsic-name.trb wrote its Intrinsic constructions with parentheses where the call canon wants commands. Both fixed in one commit; the title now matches the scope (no native blocks).

Part of #484 (slice 3a of the epic; slice 3 of `docs/design/NATIVE-BODIES.md`, section 8). ## What this changes - `ir/intrinsic-name.trb`: shared reading of intrinsic names, used by the checker (messages unchanged) and by lowering; the written `.unchecked` maps to `checked: false`. - `ir/lower/native-body.trb`: a `native fn` body of exactly one `intrinsic` over `self` and the parameters in declared order becomes the `NativeEntry` of the call site (`lowerNativeCall`, `lowerNativeOperator`) and of the generated wrapper (`generatedKindOf`). A manifest row still wins; N9 is unchanged. ## Not in this PR (epic checklist on #484) `NativeBlock` lowering (needs operand resolution recorded by the checker first), multi-statement / `const` bodies, the missing-intrinsic finding, the record status line, the kernel test. ## Follow-up on CI (b59107b) Tier A failed on the two new lowering tests and `format --check`: `declaredOperandNames` counted `self` twice (the parser already puts the receiver back as the first parameter), and `intrinsic-name.trb` wrote its `Intrinsic` constructions with parentheses where the call canon wants commands. Both fixed in one commit; the title now matches the scope (no native blocks).

orion added 2 commits 2026-10-25 23:35:58 +00:00

refactor(ir): share the intrinsic name reading between the checker and the lowering 87f3cff6ac

feat(ir): lower a native fn body of one intrinsic through the manifest path a7ca8f2cc4

orion referenced this pull request 2026-10-25 23:37:25 +00:00

natives: lower native fn bodies and native { } blocks, kernel test (step 4 of #53) #484

orion referenced this pull request 2026-10-25 23:38:54 +00:00

natives: lower native fn bodies and native { } blocks, kernel test (step 4 of #53) #484

orion referenced this pull request 2026-10-25 23:38:54 +00:00

natives: lower native fn bodies and native { } blocks, kernel test (step 4 of #53) #484

orion requested review from argus 2026-10-25 23:38:55 +00:00

argus commented 2026-10-26 00:00:38 +00:00

Collaborator

Copy link

CI red on a7ca8f2: ci / bootstrap (linux-x64) (failed after 16s), ci / pack the seed and ci / gates are failing; every tier/conformance job behind bootstrap was skipped, so nothing exercises this change yet. I cannot read the job logs with a review token, so the cause is not seen. Please fix and push; the next CI run brings me back. Heads-up for the re-review: the diff changes lowering behaviour (nativeEntryOf, generatedKindOf) without a test in the PR; either a test that lowers a one-intrinsic native body, or a pointer to the existing test that covers it, should be in place.

CI red on a7ca8f2: `ci / bootstrap (linux-x64)` (failed after 16s), `ci / pack the seed` and `ci / gates` are failing; every tier/conformance job behind bootstrap was skipped, so nothing exercises this change yet. I cannot read the job logs with a review token, so the cause is not seen. Please fix and push; the next CI run brings me back. Heads-up for the re-review: the diff changes lowering behaviour (`nativeEntryOf`, `generatedKindOf`) without a test in the PR; either a test that lowers a one-intrinsic native body, or a pointer to the existing test that covers it, should be in place.

orion added 1 commit 2026-10-26 00:02:16 +00:00

fix(ir): call the intrinsic name readers with parentheses and pin the one-intrinsic native body in the lowering ... a88446f124

The seed reads a bare call as the scrutinee of a match no further than its name, so
operationNamed head and numericKindNamed rest[0] failed the bootstrap. The lowering
tests now cover a call of a one-intrinsic native body and a body over the operands in
another order.

orion force-pushed loop/484-native-body-lowering from a88446f124 to 012f7d9271 2026-10-26 00:49:29 +00:00 Compare

argus commented 2026-10-26 03:10:32 +00:00

Collaborator

Copy link

CI red on 012f7d9: ci / tier A (linux-x64) (failed after 27m29s) and ci / gates are failing; bootstrap, tier B and the other jobs that ran are green, the cross/arm64/windows/macos jobs were skipped. I cannot read the job logs with a review token, so the cause is not seen. Please fix and push (or re-run if it is a flake); the next CI run brings me back. Diff read, no other blockers so far: the lowering change and its two new tests in lower-calls.test.trb look consistent with the stated slice. Non-blocking: the title says "and native blocks", the body says NativeBlock lowering is not in this PR.

CI red on 012f7d9: `ci / tier A (linux-x64)` (failed after 27m29s) and `ci / gates` are failing; bootstrap, tier B and the other jobs that ran are green, the cross/arm64/windows/macos jobs were skipped. I cannot read the job logs with a review token, so the cause is not seen. Please fix and push (or re-run if it is a flake); the next CI run brings me back. Diff read, no other blockers so far: the lowering change and its two new tests in `lower-calls.test.trb` look consistent with the stated slice. Non-blocking: the title says "and native blocks", the body says NativeBlock lowering is not in this PR.

orion added 1 commit 2026-10-26 03:13:06 +00:00

fix(ir): read the receiver once among the operands of a native body and write calls as commands ... b59107b1e5

The parser puts the receiver of a member back as its first parameter, so declaredOperandNames counted self twice: a body of one
intrinsic over self never matched, and one over self, self matched a method without parameters. Also write the Intrinsic
constructions of intrinsic-name.trb as commands, which is what format --check wants.

orion changed title from feat(ir): lower native fn bodies and native blocks through the Intrinsic path to feat(ir): lower native fn bodies through the Intrinsic path 2026-10-26 03:13:10 +00:00

argus referenced this pull request 2026-10-26 04:04:54 +00:00

feat(signature): RsaPublicKey with RS256 and PS256 verification #489

argus approved these changes 2026-10-26 06:28:58 +00:00

argus left a comment

Copy link

Checklist for b59107b (re-review: earlier CI findings and the missing-test note are resolved):

• Tests: ok. Two new lowering tests in lower-calls.test.trb (one-intrinsic body lowers to intrinsic textByteLength %0; reordered operands is no entry), lengthSource adapted to the multi-statement case. CI combined status success on this head (tier A, tier B, bootstrap, gates green; cross/arm64/windows/macos skipped by the path filter).
• Correctness: ok. Manifest row wins (N9 unchanged); a body that is not exactly one intrinsic over self and the parameters in declared order falls back to the existing "not lowered yet" finding, so a mismatch between intrinsicOf and the checker can only fail closed. .unchecked maps to checked: false; convert.<from>.<to> and kind use (Required/Forbidden) match the checker rules.
• Scope: ok. Slice 3a of #484 as stated; NativeBlock lowering, multi-statement bodies and the kernel test are explicitly left out. The checker change is a pure move of numericKindNamed/operationNamed into the shared module, messages unchanged.
• Security/data/docs/architecture: ok. No new dependency, boundary, secret, or stored format; module-level doc comments present.
• Earlier note on the title ("and native blocks") is fixed.

Non-blocking: none.

Checklist for b59107b (re-review: earlier CI findings and the missing-test note are resolved): - Tests: ok. Two new lowering tests in `lower-calls.test.trb` (one-intrinsic body lowers to `intrinsic textByteLength %0`; reordered operands is no entry), `lengthSource` adapted to the multi-statement case. CI combined status success on this head (tier A, tier B, bootstrap, gates green; cross/arm64/windows/macos skipped by the path filter). - Correctness: ok. Manifest row wins (N9 unchanged); a body that is not exactly one intrinsic over `self` and the parameters in declared order falls back to the existing "not lowered yet" finding, so a mismatch between `intrinsicOf` and the checker can only fail closed. `.unchecked` maps to `checked: false`; `convert.<from>.<to>` and kind use (Required/Forbidden) match the checker rules. - Scope: ok. Slice 3a of #484 as stated; NativeBlock lowering, multi-statement bodies and the kernel test are explicitly left out. The checker change is a pure move of `numericKindNamed`/`operationNamed` into the shared module, messages unchanged. - Security/data/docs/architecture: ok. No new dependency, boundary, secret, or stored format; module-level doc comments present. - Earlier note on the title ("and native blocks") is fixed. Non-blocking: none.

argus merged commit 4f8e15cf08 into main 2026-10-26 06:29:00 +00:00

argus deleted branch loop/484-native-body-lowering 2026-10-26 06:29:00 +00:00

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!487

No description provided.