feat(emitter): #46 — row-chunk monster sheet modules behind a same-named facade#70
Merged
Merged
Conversation
…med facade V8 fatally allocates ~2x a module's bytes to UTF-16-decode its source at import time, so a transpiled monster sheet crashed node before compute() ever ran (the real A-2 Debt sheet: ~305MB module -> 609,447,784-byte fatal alloc -> no converged A-2 base obtainable). The emitter now rotates any sheet module that crosses --max-module-mb (default 64, single-token --max-module-mb=N form, 0 disables) into <Sheet>.partNNN.mjs modules behind a same-named facade: - ONE logical compute(): the facade exports the same SHEET_NAME / SHEET_DEPENDENCIES / compute contract and invokes the parts in emission order — engine.js, per-sheet-eval, cone emission, cell-exprs and every other consumer import the sheet by the exact same path as before - splits happen only at statement boundaries in the linear regions; the intra-sheet convergence loop is INDIVISIBLE (one checkpoint before it, none inside) - a sheet under the cap stays a single byte-identical file (golden master unaffected); stale part files from a previous differently-sized build are swept so sheets/*.mjs sweeps can't pick up stale cells - per-sheet-eval's OFFSET/INDIRECT scan now follows the facade's part imports (it read only the facade and would have approved GT-seed scoping for exactly the monster sheets most likely to use OFFSET — red pre-fix in the new test) Real-model gate (A-2, the #46 crasher): Debt emits as 5 parts (4x64MB + 53MB) and imports in 2.4s on a DEFAULT heap; the full 21-sheet engine imports in 8.9s. Synthetic regression test-row-chunked-modules.mjs: 19 asserts — facade/ part layout + size bounds, convergence-loop indivisibility, GT reproduction, split-vs-single IDENTICAL values, cell-exprs surface parity, stale-part sweep, and the eval-scan companion (negative-controlled via stash). cargo tests 29/29, full npm test green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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.
What
Closes #46. V8 fatally allocates ~2× a module's bytes to UTF-16-decode its source at import, so a transpiled monster sheet crashed node before
compute()ever ran — A-2's ~305MB Debt module was the observed609,447,784-bytefatal alloc that made a converged A-2 base unobtainable.The emitter now rotates any sheet module crossing
--max-module-mb(default 64;--max-module-mb=N; 0 disables) into<Sheet>.partNNN.mjsmodules behind a same-named facade:compute()— the facade exports the sameSHEET_NAME/SHEET_DEPENDENCIES/computecontract and invokes parts in emission order. engine.js, per-sheet-eval, cone emission, cell-exprs: zero consumer changes.sheets/*.mjssweeps can't pick up stale cells.Gates
test-row-chunked-modules.mjs, 19 asserts: facade/part layout + size bounds, convergence-loop indivisibility, GT reproduction through the parts, split-vs-single identical values, cell-exprs surface parity, stale-part sweep, eval-scan companion (negative-controlled via stash — red pre-fix).npm testgreen.🤖 Generated with Claude Code