[01/06] feat(core): add bounded functions exec foundations - #3
Draft
HeavenllyDemon wants to merge 2 commits into
Draft
[01/06] feat(core): add bounded functions exec foundations#3HeavenllyDemon wants to merge 2 commits into
HeavenllyDemon wants to merge 2 commits into
Conversation
HeavenllyDemon
force-pushed
the
feat/functions-exec-01-foundations
branch
from
August 18, 2026 13:15
bf1b550 to
13c96f1
Compare
HeavenllyDemon
changed the title
HeavenllyDemon
added a commit
that referenced
this pull request
Aug 20, 2026
USER LIVE-GATE FIX #4, research half: does LOK paint an infinite grid past a spreadsheet's used range, the way Excel/Calc's own UI does? Answered empirically against our real vendored LOK pin, not assumed — OfficeHelperLiveTests.testGateXlsxTilesPastTheUsedRangeEmpiricalInfinite GridProbe requests gate.xlsx tiles at the used-range edge and ~20 tiles beyond it: both come back as a clean, gridded, non-blank canvas (PNG dump), unchanged all the way out. paintPartTile genuinely renders empty cells there; nothing needed faking client-side. Implementation: OfficeTileCanvasView.effectiveExtentTwips extends the scrollable bound by 2 extra screens per axis, for spreadsheets only (isSpreadsheet, read once from the open document's type — presentations/ documents have real fixed page bounds and were never probed for this). One computed property feeds BOTH the scroll clamp (clampedOriginX/Y) and performSubscribe's churn-audit skip-check (officeClampViewportToDocumentExtent's call site) so they cannot disagree — the trap being guarded against: if scrolling reached further than the skip-check's own clamp, the margin would be scrollable but its tiles would never actually get requested (placeholders forever, just moved past the new margin instead of past sizeTwips). That widened skip-check interacts with fix #3's whole-document residency sweep at a small, fully-resident spreadsheet's own edge: the fixed subscribeMarginPoints overscan now genuinely touches real, never- prefetched margin tiles, so a resident document's post-fill scroll near its edge now costs exactly one further subscribe (a one-time margin warm), not the prior zero. testResidentDocumentIsPrefetchedWholeIn... NoFurtherRequests is amended (not weakened) into a two-phase pin proving exactly that: one warm, then genuine silence once the touched slice is also cached — the "no chatter for zero benefit" invariant's spirit holds, only its letter moved.
HeavenllyDemon
added a commit
that referenced
this pull request
Aug 25, 2026
…rsor position before typing Coordinator review, Important #2 and #3, addressed together in writeOneCellOnDedicatedThread since both land in the same function: #2 — formulaKeyEvent(for:) was called inside the typing loop itself, so an unmapped character (any non-ASCII, e.g. =A1&"café") threw AFTER already posting every character before it, leaving a real, uncommitted, partial formula in Calc's own edit mode on a document a human may have open. Fixed by pre-validating every character in a pure pass before the first keystroke; a new SaveError.unsupportedFormulaCharacter replaces the old writeVerificationFailed mislabel for this path. #3 — the old positioning check trusted GoToCell's own best-effort dance and moved straight to typing; on a non-empty bystander cell, a GoToCell that never landed would leave the cursor on the WRONG cell, which already has content, so the lenient post-write check read non-empty and passed. Closed via getCommandValues(".uno:CellCursor") — a live probe (OfficeHelperLiveTests.testProbeInvestigatesWhetherCellAddressCallbacks...) first found the obvious candidate (LOK_CALLBACK_CELL_ADDRESS/CELL_CURSOR, raw types 34/17) never fires for a GoToCell-driven move at all, only real mouse/key input; getCommandValues is a synchronous query instead, called within the same dedicated-thread closure that positioned the cursor, so it is immune to the async-callback-staleness class this file's own GoToCell docs already cover. A (column,row) mismatch now throws SaveError.positionVerificationFailed before any keystroke is posted. Both fixes deletion-red verified live: temporarily reverted each guard, confirmed real failures (partial formula characters typed; every correctly positioned write wrongly refused), then reverted. Full OfficeSheetsCommandTests class: 19/19 after the fix.
HeavenllyDemon
added a commit
that referenced
this pull request
Aug 25, 2026
…s it (red/green)
The original flake could not be summoned (2 full suite passes under saturating load
never hit it), so it was injected instead — deletion-red discipline inverted: rather
than breaking a mechanism to watch a drill go red, force the FAILURE and watch the FIX
go green.
RED (forced pass-2 `body` positioning nil, retry disabled): 6 failures, and the
refusal text reproduced the original full-suite failure BYTE FOR BYTE — "body
failed after title in this SAME set_text call already applied: slide 2 in <id>
has no body placeholder — nothing was written." Same assertion count as the real
failure.
GREEN (identical force, retry enabled): Executed 1 test, with 0 failures — AND exactly
one "re-posting Escape+Tab once" evidence line.
That evidence line is the load-bearing half: it proves the forced miss still fired and
the RETRY absorbed it, rather than the probe having silently stopped working — which
would be this arc's own defect class #3 (the self-restoring instrument) reappearing
inside the proof of a fix for it.
Probe reverted, tree confirmed byte-identical. Record kept inline at the call site,
including the explicit non-overclaim: this proves the retry absorbs a single discrete
loss; that it cures the ORIGINAL flake is argued from the measured distribution, not
from a reproduced-and-cured instance.
Claude-Session: https://claude.ai/code/session_019pqCwL5mTcMEPHer8TK5D8
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First PR in the ordered functions.exec stack. Adds private, cross-platform protocol, NDJSON framing, retention, and worker-helper foundations only. No tool registration, daemon, engine, provider, persistence, or client surface is changed.
Verification:
Full Core TypeScript still has six pre-existing approvals.test.ts diagnostics; nested Seatbelt workflow execution is blocked by the enclosing host sandbox.