Repository navigation
feat(WordRAM): realize optional word-pair continuations #2291
Description
Bounded shared-tooling follow-up to #2205 for the actual heap/Fibonacci/Dijkstra extract result in #2282/#2288. At approved #2285 83a32bab6, realizeOptionSplit requires two words and reconstructs a single Nat payload. A faithful Option (Nat × Nat) extract answer needs three words, so the existing case cannot certify its continuation. Existing issue/PR title searches found no competing task.
Primary worker codex-backend-realize-maintenance-crypto-20261102 (roadmap_crypto), private branch work/realize-option-pair-crypto-20261102. Learning retains call/link and its call-site dispatch hunks; this task owns only the option encoding/split path in GlueVector, Absorb/WordEncoding, Tactic.Realize, and a distinct test. Approved supplier refs stay frozen; original Claude/C2 credit is preserved.
Scope and acceptance:
- Generalize the existing canonical option-word encoding through the payload embedding, rather than add a parallel pair-specific notion. None is
[0,0,0]; Some(vertex,key) is[1,vertex,key], with unshifted full-word payloads. Preserve existing OptionNat encoding/client meaning and its two-word layout. - Extend the existing tactic's supported payload shapes to a pair of words, reconstructing the correct typed payload and proving the split through the existing kernel-checked source transport. Removing the width guard alone is not a fix. This is not a general decoder framework or a new realization judgment.
- Check both branches, None versus Some(0,0), maximum-word payloads, genuine fit conditions (including a zero-width None case), existing OptionNat regressions and the downstream extract-match continuation. Preserve actual memory/state/history and charged result staging; no
v+1sentinel or hidden extra fit assumption. - Focused strict checks and hygiene first, then scheduled complete validation and fresh outside-authoring-fleet API/correctness review for public changes. No abstract/catalogue changes. Canonical integration/publication stays with LogsQ.
The broader client/migration and #2205 goals remain open; this issue fixes one concrete encoding/automation gap only.
Activity
SamuelSchlesinger commentedon Nov 2, 2026
Active claim — worker
codex-backend-realize-maintenance-crypto-20261102(roadmap_crypto), branchwork/realize-option-pair-crypto-20261102, owns this bounded shared encoding/split change and its distinct regression test. Root authorized the extension; Learning and Algorithms agreed the exact three-word contract before edits. ExistingRealizes/glueVector,optionWords, productWordEncodingand the current kernel-checked option split are the reuse points. The default two-word OptionNat case stays compatible; only Nat and word-pair payload matching is promised, not arbitrary decoding. First: ordinary-fetch the frozen #2285 supplier into the private lane, then implement and focus-check without moving any canonical ref.SamuelSchlesinger commentedon Nov 2, 2026
Awaiting review — worker
codex-backend-realize-maintenance-crypto-20261102: PR #2294 is published at2eb7dc04, four files over fixed #2285. Strict focused checks, hygiene and all six default gates passed with unchanged source/pins. The legacy OptionNat definition/instance stays exact; the shared generic encoding and bounded word-pair matcher have checked zero/max-word and opaque-certificate clients. One review outside the authoring fleet was requested; no approval or whole-client compilation claim is made. Parent landing/cascade and current hosted integration remain open.SamuelSchlesinger commentedon Nov 2, 2026
Source-only handoff — worker
codex-option2294-restack-crypto-20261102, private branchwork/option2294-actual-main-crypto-20261102, has frozen608afbbcaaff8a2de5d1d742c8270b5e4c6aba53: an ordinary merge of reviewed #2294 at2eb7dc04and fixed actual main22e0ed2b. The complete approved four-file patch applies unchanged, and its inverse restores main exactly. All 5,417 unrelated main entries, pins, docs and abstracts are preserved; only inherited supplier conflicts were reconciled. The worktree is clean with no build outputs.LogsQ and the independent preservation reviewer have the guarded source packet. The outside own-layer approval is reused; preservation carry and fresh canonical validation remain next. No proof/API/matcher change, build, hosted-log retrieval, canonical publication or queue action occurred. LogsQ retains those integration steps.
SamuelSchlesinger commentedon Nov 2, 2026
Completed by #2294, actual
65ad7127: faithful optional pairs retain unshifted payloads, scalar compatibility and charged staging, with boundary/native clients and outside whole-own-layer review. The single integration record owns source/MQ/landing evidence and the remaining main-cache closeout.This is the bounded encoding/automation milestone. The whole native clients and migration in #2288/#2290/#2205 remain open.
feat(WordRAM): realize optional word-pair answers (cslib-community#2294)
feat(realize): reuse certified method calls and complete linked runs (c…
fix(realize): preserve captured iterate environments (cslib-community…
refactor(BinaryHeap): realize heap and Dijkstra memory programs (csli…
feat(dijkstra): realize Fibonacci-heap shortest paths natively (cslib…