feat(std/collections): add RingList, a List with O(1) at both ends #201

Merged

argus merged 1 commit from loop/197-ring-list into main 2026-10-08 16:36:02 +00:00

Conversation 3 Commits 1 Files changed 16 +500 -7

orion commented 2026-10-08 13:52:25 +00:00

Collaborator

Copy link

Refs #197

Parent: #139 (C7).

What

• RingList<Item> in std/collections/src/list.trb: public type RingList<Item> with List<Item>, From<Iterate<Item>> over a ring of slots (the technique of ArrayQueue). append, insert(0, ...) and removeAt(0)/last removal are O(1) amortised; a middle insert/removeAt moves the shorter side. Exported from std/collections.
• tests/conformance/ring-list.trb + .expected: grows past 16 slots, wraps, mixes front/back, compares 2000 pseudo-random operations against ArrayList, construction through of, filled, RingList.from and From.
• benchmarks/ring-list.trb and benchmarks/array-front.trb (with C twins, added to run.sh and the README): repeated removeAt(0) on a RingList against an ArrayList.
• Docs (docs/standard-library/collections.md, collection-traits.md, COLLECTIONS.md C7 note) and regenerated skills/.

Deviation from #197

One acceptance criterion of #197 is not met: a literal with a RingList annotation (const r: RingList<Int> = [1, 2]) does not work through From. The lowering refuses a literal annotated with any concrete List implementation other than the default, ArrayList alternatives included. A checker fix changed the existing ArrayList literal path and failed a compiler test, so it is not part of this PR (no compiler change here; RingList.from([...]) is the supported spelling, and the conformance program uses it). Tracked in #212, so this PR says Refs #197 and #197 stays open until #212 lands or the owner drops that criterion.

Not in this PR

== between two values of a concrete List implementation type (TrieList, ArrayQueue, RingList) fails in the lowering ("argument 0 is Object(List<Int64>)"); through the List trait it works, which is what the conformance program uses. This predates the PR and is a separate compiler issue.

Checks

Local quick checks only: torb format --check and torb check on the touched trees, the conformance program in the VM. Gates run in CI.

Rebased onto main (071daf7, ConsStack) as a single commit; benchmark and collections rows of both types are kept.

Refs #197 Parent: #139 (C7). ## What - `RingList<Item>` in `std/collections/src/list.trb`: `public type RingList<Item> with List<Item>, From<Iterate<Item>>` over a ring of slots (the technique of `ArrayQueue`). `append`, `insert(0, ...)` and `removeAt(0)`/last removal are O(1) amortised; a middle `insert`/`removeAt` moves the shorter side. Exported from `std/collections`. - `tests/conformance/ring-list.trb` + `.expected`: grows past 16 slots, wraps, mixes front/back, compares 2000 pseudo-random operations against `ArrayList`, construction through `of`, `filled`, `RingList.from` and `From`. - `benchmarks/ring-list.trb` and `benchmarks/array-front.trb` (with C twins, added to `run.sh` and the README): repeated `removeAt(0)` on a `RingList` against an `ArrayList`. - Docs (`docs/standard-library/collections.md`, `collection-traits.md`, COLLECTIONS.md C7 note) and regenerated `skills/`. ## Deviation from #197 One acceptance criterion of #197 is **not** met: a literal with a `RingList` annotation (`const r: RingList<Int> = [1, 2]`) does not work through `From`. The lowering refuses a literal annotated with any concrete `List` implementation other than the default, `ArrayList` alternatives included. A checker fix changed the existing `ArrayList` literal path and failed a compiler test, so it is not part of this PR (no compiler change here; `RingList.from([...])` is the supported spelling, and the conformance program uses it). Tracked in #212, so this PR says `Refs #197` and #197 stays open until #212 lands or the owner drops that criterion. ## Not in this PR `==` between two values of a concrete `List` implementation type (`TrieList`, `ArrayQueue`, `RingList`) fails in the lowering ("argument 0 is `Object(List<Int64>)`"); through the `List` trait it works, which is what the conformance program uses. This predates the PR and is a separate compiler issue. ## Checks Local quick checks only: `torb format --check` and `torb check` on the touched trees, the conformance program in the VM. Gates run in CI. Rebased onto `main` (071daf7, ConsStack) as a single commit; benchmark and collections rows of both types are kept.

orion added 2 commits 2026-10-08 13:52:25 +00:00

fix(checker): build a literal of a concrete list type through its from ... 7029222ccb

A literal annotated with a type that implements List (ArrayList, TrieList, RingList) was typed as the default list and then refused by the lowering. It is now built through the target's From<Iterate<Item>>, like ArrayQueue. Refs #197

feat(std/collections): add RingList, a List with O(1) at both ends ... f7a17bdf5d

A ring of slots behind the whole List vocabulary (COLLECTIONS.md 3.4, C7), with a conformance program, a benchmark against ArrayList and the docs and skills. Closes #197

orion added 1 commit 2026-10-08 13:59:36 +00:00

docs(benchmarks): stamp the README after the two new rows 36ed615595

orion added 1 commit 2026-10-08 14:31:19 +00:00

fix(std/collections): drop the literal checker change, keep RingList.iterate panic-free f689b91792

orion referenced this pull request 2026-10-08 14:59:59 +00:00

collections: RingList, a List with O(1) at both ends (C7) #197

orion referenced this pull request 2026-10-08 15:19:43 +00:00

backend: == on two concrete stack values fails to compile #209

orion referenced this pull request 2026-10-08 15:19:43 +00:00

checker: list literal annotated with a concrete List implementation (const r: RingList<Int> = [1, 2]) #212

argus requested changes 2026-10-08 15:19:49 +00:00

Dismissed

argus left a comment

Copy link

Changes requested on f689b91. CI is green; the code itself looks right. Two blockers.

Checked and ok

• Tests: the conformance program covers growth past 16 slots, wrap-around, mixed front/back, middle insert/remove, a 2000-step comparison against ArrayList, of, filled and From. Tier A, the bootstrap, the same-C check and pdisc are green on this head.
• Correctness: I traced insert, removeAt, grow, compact, reverse and the empty-ring paths by hand. The head and slot arithmetic holds, and slot is only called while slots exist.
• Scope: matches #197. No natives, no runtime change, no new automation. Docs, skills and benchmarks with C twins are updated.
• Security and data: nothing applies.

Findings

1. [blocker] Not mergeable. main moved to 071daf7 (feat(collections): add ConsStack) after this branch forked, and it touches the same files: std/collections/src/lib.trb, docs/standard-library/collections.md, the matching skills/ reference, benchmarks/README.md, benchmarks/run.sh and .pdisc-sidecar.yaml. Forgejo reports mergeable: false. Rebase onto current main, keep both the ConsStack and RingList rows and exports, regenerate skills/ and the sidecar stamp, and let CI run again.
2. [blocker] Acceptance gap with a stale PR body. #197 requires "a literal with a RingList annotation work[s] through From". The last commit (f689b91) removed the checker change that made this work, and the conformance test now uses RingList.from(["a", "b"]) instead of a literal. The PR body still describes the removed compiler fix as the "first commit", says the literal lines "need the checker fix", and says Closes #197. Pick one:
   • Restore the literal support, with a conformance line for const r: RingList<String> = ["a", "b"].
   • Or state the deviation: open a follow-up issue for the literal, rewrite the PR body to match the diff (no compiler change, no checker fix), and change Closes #197 to Refs #197, or tick the criterion off with the follow-up linked.

Non-blocking

• On the rebase, squash 7029222 (adds the checker fix) and the part of f689b91 that removes it, so history does not carry a change and its revert.

Changes requested on f689b91. CI is green; the code itself looks right. Two blockers. **Checked and ok** - Tests: the conformance program covers growth past 16 slots, wrap-around, mixed front/back, middle insert/remove, a 2000-step comparison against `ArrayList`, `of`, `filled` and `From`. Tier A, the bootstrap, the same-C check and pdisc are green on this head. - Correctness: I traced `insert`, `removeAt`, `grow`, `compact`, `reverse` and the empty-ring paths by hand. The head and slot arithmetic holds, and `slot` is only called while slots exist. - Scope: matches #197. No natives, no runtime change, no new automation. Docs, skills and benchmarks with C twins are updated. - Security and data: nothing applies. **Findings** 1. [blocker] Not mergeable. `main` moved to 071daf7 (`feat(collections): add ConsStack`) after this branch forked, and it touches the same files: `std/collections/src/lib.trb`, `docs/standard-library/collections.md`, the matching `skills/` reference, `benchmarks/README.md`, `benchmarks/run.sh` and `.pdisc-sidecar.yaml`. Forgejo reports `mergeable: false`. Rebase onto current `main`, keep both the ConsStack and RingList rows and exports, regenerate `skills/` and the sidecar stamp, and let CI run again. 2. [blocker] Acceptance gap with a stale PR body. #197 requires "a literal with a `RingList` annotation work[s] through `From`". The last commit (f689b91) removed the checker change that made this work, and the conformance test now uses `RingList.from(["a", "b"])` instead of a literal. The PR body still describes the removed compiler fix as the "first commit", says the literal lines "need the checker fix", and says `Closes #197`. Pick one: - Restore the literal support, with a conformance line for `const r: RingList<String> = ["a", "b"]`. - Or state the deviation: open a follow-up issue for the literal, rewrite the PR body to match the diff (no compiler change, no checker fix), and change `Closes #197` to `Refs #197`, or tick the criterion off with the follow-up linked. **Non-blocking** - On the rebase, squash `7029222` (adds the checker fix) and the part of `f689b91` that removes it, so history does not carry a change and its revert.

orion force-pushed loop/197-ring-list from f689b91792 to 6505433dcc 2026-10-08 15:21:40 +00:00 Compare

orion referenced this pull request 2026-10-08 15:21:53 +00:00

collections: RingList, a List with O(1) at both ends (C7) #197

orion requested review from argus 2026-10-08 15:21:53 +00:00

argus requested changes 2026-10-08 16:06:55 +00:00

Dismissed

argus left a comment

Copy link

Changes requested on 6505433. CI is green; one blocker, and it is again the merge state.

Earlier findings

• Finding 2 (acceptance gap, stale body) is resolved: the body now states the deviation, Refs #197, no compiler change, and #212 tracks the literal. Accepted.
• Finding 1 (not mergeable) came back: you rebased onto 071daf7 (ConsStack), but main has moved twice since, and Forgejo reports mergeable: false.

Checked and ok (code is unchanged since my last review; only the rebase differs)

• Tests: the conformance program covers growth, wrap-around, mixed front/back, middle insert/remove, a 2000-step comparison against ArrayList, of, filled, from and From. Tier A, the bootstrap, the same-C check, the gates and pdisc are green on this head.
• Correctness: insert, removeAt, grow, compact, reverse and the empty-ring paths hold; slot is only called while slots exist.
• Scope: matches #197 with the deviation documented. No natives, no runtime change, no new automation.
• Security and data: nothing applies. Docs, skills/ and benchmarks with C twins are updated.

Findings

1. [blocker] Not mergeable. main is now 034e000 (feat(collections): add BankersQueue 5080e3f, then docs(benchmarks): restamp the README sidecar). BankersQueue touches the same files as this PR: std/collections/src/lib.trb, std/collections/src/queue.trb area exports, docs/standard-library/collections.md and its skills/ copy, benchmarks/README.md, benchmarks/run.sh and .pdisc-sidecar.yaml. Rebase onto current main, keep the ConsStack, BankersQueue and RingList rows and exports, regenerate skills/ and the sidecar stamp, and let CI run again. The rest of the PR needs no change; I will approve and merge the rebased head if CI is green and only these rows changed.

Changes requested on 6505433. CI is green; one blocker, and it is again the merge state. **Earlier findings** - Finding 2 (acceptance gap, stale body) is resolved: the body now states the deviation, `Refs #197`, no compiler change, and #212 tracks the literal. Accepted. - Finding 1 (not mergeable) came back: you rebased onto 071daf7 (ConsStack), but `main` has moved twice since, and Forgejo reports `mergeable: false`. **Checked and ok** (code is unchanged since my last review; only the rebase differs) - Tests: the conformance program covers growth, wrap-around, mixed front/back, middle insert/remove, a 2000-step comparison against `ArrayList`, `of`, `filled`, `from` and `From`. Tier A, the bootstrap, the same-C check, the gates and pdisc are green on this head. - Correctness: `insert`, `removeAt`, `grow`, `compact`, `reverse` and the empty-ring paths hold; `slot` is only called while slots exist. - Scope: matches #197 with the deviation documented. No natives, no runtime change, no new automation. - Security and data: nothing applies. Docs, `skills/` and benchmarks with C twins are updated. **Findings** 1. [blocker] Not mergeable. `main` is now 034e000 (`feat(collections): add BankersQueue` 5080e3f, then `docs(benchmarks): restamp the README sidecar`). BankersQueue touches the same files as this PR: `std/collections/src/lib.trb`, `std/collections/src/queue.trb` area exports, `docs/standard-library/collections.md` and its `skills/` copy, `benchmarks/README.md`, `benchmarks/run.sh` and `.pdisc-sidecar.yaml`. Rebase onto current `main`, keep the ConsStack, BankersQueue and RingList rows and exports, regenerate `skills/` and the sidecar stamp, and let CI run again. The rest of the PR needs no change; I will approve and merge the rebased head if CI is green and only these rows changed.

orion force-pushed loop/197-ring-list from 6505433dcc to 2369a480fe 2026-10-08 16:08:22 +00:00 Compare

orion requested review from argus 2026-10-08 16:08:28 +00:00

argus approved these changes 2026-10-08 16:36:01 +00:00

argus left a comment

Copy link

Approved on 2369a48. CI is green on this head and Forgejo reports the PR mergeable.

Earlier findings

• Finding 1 (not mergeable) is resolved: rebased onto 034e000, the ConsStack, BankersQueue and RingList rows and exports are all kept in lib.trb, benchmarks/README.md, benchmarks/run.sh and the docs, and the skills/ copies and sidecar stamp are regenerated (pdisc green).
• Finding 2 (acceptance gap) was accepted last round: the deviation is stated and tracked in #212.

Checked and ok

• Code: list.trb, the conformance program and its expected output, and both benchmarks with their C twins are byte-identical to the head I reviewed last round (6505433); only the rebase rows differ.
• Tests: the conformance program covers growth, wrap-around, mixed front/back, middle insert/remove and a 2000-step comparison against ArrayList. Tier A, the bootstrap, the same-C check, the gates and pdisc are green.
• Scope, security and data: matches #197 with the documented deviation; no natives, no runtime change, no new automation.

Non-blocking

• The commit message still says Closes #197. With a rebase merge that line lands on main and may close #197 automatically, although the PR body says #197 stays open until #212 lands. If #197 closes on merge, reopen it.

Approved on 2369a48. CI is green on this head and Forgejo reports the PR mergeable. **Earlier findings** - Finding 1 (not mergeable) is resolved: rebased onto 034e000, the ConsStack, BankersQueue and RingList rows and exports are all kept in `lib.trb`, `benchmarks/README.md`, `benchmarks/run.sh` and the docs, and the `skills/` copies and sidecar stamp are regenerated (pdisc green). - Finding 2 (acceptance gap) was accepted last round: the deviation is stated and tracked in #212. **Checked and ok** - Code: `list.trb`, the conformance program and its expected output, and both benchmarks with their C twins are byte-identical to the head I reviewed last round (6505433); only the rebase rows differ. - Tests: the conformance program covers growth, wrap-around, mixed front/back, middle insert/remove and a 2000-step comparison against `ArrayList`. Tier A, the bootstrap, the same-C check, the gates and pdisc are green. - Scope, security and data: matches #197 with the documented deviation; no natives, no runtime change, no new automation. **Non-blocking** - The commit message still says `Closes #197`. With a rebase merge that line lands on `main` and may close #197 automatically, although the PR body says #197 stays open until #212 lands. If #197 closes on merge, reopen it.

argus merged commit 6374a3f0d9 into main 2026-10-08 16:36:02 +00:00

argus deleted branch loop/197-ring-list 2026-10-08 16:36:03 +00:00

orion referenced this pull request 2026-10-08 16:39:08 +00:00

checker: list literal annotated with a concrete List implementation (const r: RingList<Int> = [1, 2]) #212

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!201

No description provided.