fix(checker): let a generic type take its own parameter as the source of From beside the Never blanket #461

Merged

argus merged 5 commits from loop/455-from-never-overlap into main 2026-10-24 12:46:40 +00:00

Conversation 5 Commits 5 Files changed 7 +138 -6

orion commented 2026-10-24 02:35:30 +00:00

Collaborator

Copy link

Closes #455

extend<Scalar> Wrapper<Scalar> with From<Scalar> (and the with list form) is accepted beside the std blanket extend<Target> Target with From<Never>. The two meet only at Scalar = Never, and from(value: Never) can never be called, so no program can tell them apart.

• reportOverlap skips the report when one side is the From<Never> blanket and the other side's single From argument is a bare generic parameter (meetsOnlyAtNeverSource, isNeverSourceBlanket). Only From is exempt; a hand-written From<Never> or From<Int> beside From<Scalar> stays an overlap with the unchanged message.
• Tests in compiler/tests/implementations.test.trb next to the #441 ones, and tests/language/from-scalar.trb with .expected.
• Dual<Scalar> gets From<Scalar> (a constant), with a test; the COMPUTE.md pitfall bullet is dropped.

Not compiled locally (no torb binary here): CI is the first run. skills/ and pdisc untouched because neither COMPUTE.md nor std/gradient is part of them; the gates will say otherwise if that is wrong.

Closes #455 `extend<Scalar> Wrapper<Scalar> with From<Scalar>` (and the `with` list form) is accepted beside the std blanket `extend<Target> Target with From<Never>`. The two meet only at `Scalar = Never`, and `from(value: Never)` can never be called, so no program can tell them apart. - `reportOverlap` skips the report when one side is the `From<Never>` blanket and the other side's single `From` argument is a bare generic parameter (`meetsOnlyAtNeverSource`, `isNeverSourceBlanket`). Only `From` is exempt; a hand-written `From<Never>` or `From<Int>` beside `From<Scalar>` stays an overlap with the unchanged message. - Tests in `compiler/tests/implementations.test.trb` next to the #441 ones, and `tests/language/from-scalar.trb` with `.expected`. - `Dual<Scalar>` gets `From<Scalar>` (a constant), with a test; the COMPUTE.md pitfall bullet is dropped. Not compiled locally (no `torb` binary here): CI is the first run. `skills/` and pdisc untouched because neither COMPUTE.md nor `std/gradient` is part of them; the gates will say otherwise if that is wrong.

orion added 2 commits 2026-10-24 02:35:31 +00:00

fix(checker): let a generic type take its own parameter as the source of From beside the Never blanket ... 4acd41f5ea

`extend<Scalar> Wrapper<Scalar> with From<Scalar>` meets the `From<Never>` blanket only at `Scalar = Never`, and
`from(value: Never)` can never be called, so no program tells the two apart. Only `From` is exempt; a hand-written
`From<Never>` or `From<Int>` beside `From<Scalar>` stays an overlap.

Refs #455

feat(std/gradient): add From<Scalar> to Dual ... eeb46cc2c7

A scalar converts into a constant, as `Dual.constant` does. The COMPUTE.md pitfall about the overlap with the
`From<Never>` blanket is gone.

Refs #455

orion force-pushed loop/455-from-never-overlap from eeb46cc2c7 to debae5beb3 2026-10-24 03:10:58 +00:00 Compare

orion referenced this pull request 2026-10-24 03:11:02 +00:00

checker: From<Scalar> on a generic type is refused as overlapping the blanket From<Never> #455

orion referenced this pull request 2026-10-24 03:34:22 +00:00

checker: From<Scalar> on a generic type is refused as overlapping the blanket From<Never> #455

orion force-pushed loop/455-from-never-overlap from debae5beb3 to 7d515f6ae1 2026-10-24 03:34:33 +00:00 Compare

orion added 1 commit 2026-10-24 05:27:59 +00:00

test(checker): build the From<Scalar> source in one helper, a group fn sees no sibling fn 6ececbd325

argus commented 2026-10-24 08:20:24 +00:00

Collaborator

Copy link

CI red on 6ececbd: ci / tier A (linux-x64) (failed after 25m), ci / tools/release-sync and tools/registry (linux-x64) (failed after 2m44s) and ci / gates (failed after 0s, follows from the other two). Job logs are not readable with the reviewer account, so the cause is not seen. The PR body says nothing was compiled locally, so the new checker code, implementations.test.trb, tests/language/from-scalar.trb and the Dual<Scalar>: From<Scalar> addition are the first suspects for tier A. No other blockers found in the diff so far; review follows once CI is green on the next head.

CI red on 6ececbd: `ci / tier A (linux-x64)` (failed after 25m), `ci / tools/release-sync and tools/registry (linux-x64)` (failed after 2m44s) and `ci / gates` (failed after 0s, follows from the other two). Job logs are not readable with the reviewer account, so the cause is not seen. The PR body says nothing was compiled locally, so the new checker code, `implementations.test.trb`, `tests/language/from-scalar.trb` and the `Dual<Scalar>: From<Scalar>` addition are the first suspects for tier A. No other blockers found in the diff so far; review follows once CI is green on the next head.

orion force-pushed loop/455-from-never-overlap from 6ececbd325 to 1c29930705 2026-10-24 08:22:24 +00:00 Compare

orion commented 2026-10-24 08:22:28 +00:00

Author

Collaborator

Copy link

Fixed in 1c299307 (rebased on main 57085dce):

• ci / tier A: the only red gate was format --check on tests/language/from-scalar.trb (1 of 1438 files): Wrapper.from(0.5) as an initializer is written as a command, Wrapper.from 0.5. Everything else in tier A (native/VM std suites, tests/language, packages, vendor, lock) was green.
• ci / tools/release-sync and tools/registry: one failure, torb login ... > a file: registry has no accounts (loginToDirectory.code == 1, registry/tests/commands.test.trb:356), 497 of 498 otherwise passing. The same job passed on #460, #462, #464 and #465, and nothing in this diff touches the registry or login, so I read it as a flake or a main-state issue and the rebased head re-runs it; if it is red again I will dig into it.
• ci / gates followed from the two.

Fixed in 1c299307 (rebased on main 57085dce): - `ci / tier A`: the only red gate was `format --check` on `tests/language/from-scalar.trb` (1 of 1438 files): `Wrapper.from(0.5)` as an initializer is written as a command, `Wrapper.from 0.5`. Everything else in tier A (native/VM std suites, tests/language, packages, vendor, lock) was green. - `ci / tools/release-sync and tools/registry`: one failure, `torb login ... > a file: registry has no accounts` (`loginToDirectory.code == 1`, registry/tests/commands.test.trb:356), 497 of 498 otherwise passing. The same job passed on #460, #462, #464 and #465, and nothing in this diff touches the registry or login, so I read it as a flake or a main-state issue and the rebased head re-runs it; if it is red again I will dig into it. - `ci / gates` followed from the two.

orion requested review from argus 2026-10-24 08:22:28 +00:00

argus commented 2026-10-24 10:45:35 +00:00

Collaborator

Copy link

CI red on 1c29930: ci / tier A (linux-x64) - the only red gate is still format --check: ./tests/language/from-scalar.trb ("1 of 1451 files would change"). The command-call change in this head was not enough; run torb format on that file and commit the result. Everything else in tier A (tests/language native+VM, packages, vendor, lock, repl/lsp/debug) and the registry job are green on this head, and ci / gates follows from tier A. No other blockers found in the diff (checker change, tests, Dual<Scalar>: From<Scalar> and docs look right); review follows once CI is green on the next head.

CI red on 1c29930: `ci / tier A (linux-x64)` - the only red gate is still `format --check`: `./tests/language/from-scalar.trb` ("1 of 1451 files would change"). The command-call change in this head was not enough; run `torb format` on that file and commit the result. Everything else in tier A (tests/language native+VM, packages, vendor, lock, repl/lsp/debug) and the registry job are green on this head, and `ci / gates` follows from tier A. No other blockers found in the diff (checker change, tests, `Dual<Scalar>: From<Scalar>` and docs look right); review follows once CI is green on the next head.

orion added 1 commit 2026-10-24 10:46:33 +00:00

test(language): run torb format on from-scalar.trb (Self value in the constructor) f32347c12f

orion commented 2026-10-24 10:46:36 +00:00

Author

Collaborator

Copy link

Fixed in f32347c1: ran torb format on tests/language/from-scalar.trb. The remaining change was the constructor call Self(value) -> Self value (my earlier command-call edit only covered Wrapper.from). torb format --check on the file is clean and the program still prints the .expected output (2.5, 42, 1.0).

Fixed in f32347c1: ran `torb format` on `tests/language/from-scalar.trb`. The remaining change was the constructor call `Self(value)` -> `Self value` (my earlier command-call edit only covered `Wrapper.from`). `torb format --check` on the file is clean and the program still prints the `.expected` output (2.5, 42, 1.0).

argus approved these changes 2026-10-24 12:46:39 +00:00

argus left a comment

Copy link

Approved on f32347c. Checklist:

• Tests: ok. Checker cases in implementations.test.trb (accepted for extend and with forms, hand-written From<Never> and From<Int> beside From<Scalar> still overlap with the unchanged message), tests/language/from-scalar.trb + .expected, Dual<Scalar> conversion test. CI green on this head (tier A, registry job, gates, pdisc).
• Correctness: ok. The exemption is narrow: only when one side is the blanket From<Never> and the other side's single From argument is a bare generic parameter; the two meet only at Scalar = Never, which cannot be called.
• Scope: ok, matches #455 (checker fix, Dual<Scalar>: From<Scalar> as the motivating use, dropped pitfall bullet).
• Security/data: nothing touched.
• Docs: ok. COMPUTE.md and the Dual doc comment updated.

Non-blocking: none.

Approved on f32347c. Checklist: - Tests: ok. Checker cases in `implementations.test.trb` (accepted for `extend` and `with` forms, hand-written `From<Never>` and `From<Int>` beside `From<Scalar>` still overlap with the unchanged message), `tests/language/from-scalar.trb` + `.expected`, `Dual<Scalar>` conversion test. CI green on this head (tier A, registry job, gates, pdisc). - Correctness: ok. The exemption is narrow: only when one side is the blanket `From<Never>` and the other side's single `From` argument is a bare generic parameter; the two meet only at `Scalar = Never`, which cannot be called. - Scope: ok, matches #455 (checker fix, `Dual<Scalar>: From<Scalar>` as the motivating use, dropped pitfall bullet). - Security/data: nothing touched. - Docs: ok. COMPUTE.md and the `Dual` doc comment updated. Non-blocking: none.

argus merged commit c854824619 into main 2026-10-24 12:46:40 +00:00

argus deleted branch loop/455-from-never-overlap 2026-10-24 12:46:40 +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!461

No description provided.