morluto/rea

Fork 0

mirror of https://github.com/morluto/rea.git synced 2026-11-05 06:45:22 +00:00

[PR #3456] [MERGED] fix(javascript): keep overflowing timer delays unknown #5749

New issue

Closed

opened 2026-10-31 01:09:08 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 01:09:08 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/3456
Author: @HawkForge
Created: 10/14/2026
Status: ✅ Merged
Merged: 10/20/2026
Merged by: @gensokyo-dev

Base: main ← Head: fix/timer-nonfinite-delay

📝 Commits (3)

e76985f fix(javascript): keep overflowing timer delays unknown
30936f7 test(javascript): keep the IR-level overflowing timer assertion
e487a52 fix(javascript): retain JSON-safe timer facts through CLI and MCP

📊 Changes

3 files changed (+131 additions, -1 deletions)

View changed files

📝 src/domain/javascript/javascriptSemanticAnalysis.runtime_boundaries.test.ts (+17 -0)
📝 src/domain/javascript/javascriptSemanticAsyncEffects.ts (+5 -1)
➕ tests/boundary/mcp/javascriptTimerDelay.test.ts (+109 -0)

📄 Description

Summary

Fixes #3400. A numeric literal such as 1e999 becomes Infinity in the parsed AST. The timer fact producer passed it into JSON-backed properties; projecting one timer then failed the file's analysis and discarded unrelated function facts.

The existing timerDelay owner now emits the established unknown value (null) for a non-finite literal. It preserves the timer, method, binding resolution, source range and file digest. Finite values, including 1e308, remain unchanged. CLI and MCP consume this same representation; no downstream sanitization, schema relaxation or cache is added.

A public CLI/stdio-MCP regression covers six timer cases, exact unknown-vs-omitted semantics, an unrelated exported function, source SHA-256, unchanged input and equal Evidence/node identities across adapters. The redundant new graph-only test is replaced by that stronger public workflow; the domain IR test retains its distinct binding-resolution assertion. Existing graph-consumer tests still run.

Executed validation

Pinned baseline eac296a7d, Linux x64, Node 24.18.0, npm 11.16.0:

111 targeted tests passed, covering runtime boundaries, dataflow, graph resources and the real CLI/MCP worker path. Full npm run check passed; docs:check passed, including 5 consistency tests.
Integration with newer main f5d1012fa, tree b219b2d6a6d3e8c3adec070ff81028ba1da97a92: the public CLI/MCP regression passes again, and the real-package workflow retains the same 176 nodes / 4 selected rows with verified cleanup (21.868 s elapsed, 19.831 s CPU, 1,077.8 MiB peak). Main subsequently added only an unrelated Kuna test (7b2087e44), leaving production code unchanged.
The new public regression fails on exact main at the CLI success assertion. Independent real CLI runs reproduce unreadable_output / JSON numbers must be finite for both overflowing setTimeout and setInterval: main retains one module node and no timer or greet; candidate returns 14 nodes, the unknown delay and greet for each unchanged source.
Six real CLI cases on each revision cover 1e999, 1e309, 100, 1e308, the Infinity identifier and a standalone overflowing literal. Healthy controls retain identical Evidence IDs between main and candidate. Source bytes remain unchanged; all owned temporary roots are empty.
Actual public MCP SDK 2026-07-28 on unmodified installed fast-deep-equal 3.1.3, selected index.js (1,177 bytes, SHA-256 eb469e206280321a3878f2335ec98aa2104a155079d8ed83a23029098dccd215): two summary analyses, a batch of 4 selected observation rows, then full Evidence export. The retained graph has 176 semantic nodes, including source function equal, with stable parent Evidence identity. Coverage remains partial: eight unselected files and 137 semantic unknowns are explicit. This is a small real package, not a complex-application performance claim.

REA uses static source analysis; selected target files remain unchanged. The issue's original macOS/Node 26 environment was not separately exercised.

Performance and context

One CPU, 8-GiB hard / 7-GiB early process-family memory bounds, no swap. Equivalent completed package workflows include catalog discovery, two summaries, selected facts, Evidence export and shutdown:

Revision	Total elapsed	Family CPU	Peak family memory	First / repeat analysis
Main	19.856 s	18.743 s	1,075.4 MiB	9.429 / 1.090 s
Candidate	20.967 s	18.640 s	1,147.9 MiB	10.341 / 1.110 s

The repeat uses the same MCP session and immutable input; internal reuse is uninstrumented. No new cache is introduced. OS caches and shared-host load are uncontrolled; these samples establish no general speedup or precise latency effect. All completed tasks confirm cleanup and no OOM.

The six-case CLI qualification bundles take 19.859 / 24.020 s, CPU 18.237 / 21.525 s, peak 384.5 / 384.3 MiB (main/candidate). They intentionally differ in recovered facts, so their totals are not equivalent successful-workflow speed comparisons.

cl100k_base estimates: compact input catalog 44,607 / 44,607 tokens, selected analysis input 245 / 245; the two summaries cost 2,338 tokens each, selected facts 3,647, and complete captured post-connect JSON traffic 732,048 / 732,051. The 148,881-token complete Evidence file is retained locally rather than delivered inline. Captured protocol includes output schemas; these are protocol/proxy estimates, not actual model context/usage. Initialization outside capture hooks and downstream prompts are unmeasured. Source selection, summary delivery and requested property keys bound work and inline detail while preserving full Evidence access.

CI/review assessment and authorized bypass

Latest head e487a52ac: Codex review completes with no findings. No actionable latest-head finding is open. Greptile's new run is still in progress; its earlier 5/5 is not claimed as a review of this head. Fork workflow approvals were supplied, but current classification/readiness/title jobs remain queued without completed test results.

The task owner explicitly authorized bypass once independently qualified. Merge uses completed full checks, 111 targeted tests, docs checks, exact-base CLI failures, successful CLI/MCP regression, real-package analysis and newer-main integration. No queued CI or pending Greptile result is claimed as passing; subsequent results remain subject to assessment.

Post-merge CI assessment: 35 latest-head checks succeeded; shard 1 failed only the existing protobufCapture.test.ts malformed-gRPC-web public workflow at its 30-second test timeout. Exact filtered controls reproduce the same timeout on pre-change main cf5194f68 and candidate e487a52ac; the protobuf workflow is outside this diff. Other shards were cancelled or queued. No full CI pass is claimed.

Latest Greptile review is 4/5. Its outside-diff wrapped-timer cancellation lookup finding targets line 295, unchanged by this PR and already introduced on main in 6183544e8b. The only production diff here is the timerDelay finite-number guard. The separate main concern is acknowledged with source attribution; it is not a bug introduced by this PR and is left outside this fix as the task owner requested. Codex has no findings.

Author readiness

Human owner: @gensokyo-dev

The task owner explicitly delegated validation, readiness declarations and merge to this agent. These declarations are supplied under that authorization; no separate new manual human review by the original author is asserted.

I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work.
I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps.
I declare this PR ready for initial maintainer review.

Before merge

Shared producer semantics and caller identity preserved; no new authority or schema changes.
Relevant tests, real public workflows, performance and context costs verified; no local reports, binaries, credentials or dependencies in the diff.
Available latest-head CI/review state and actionable findings assessed; queued work and bypass authorization recorded above.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/morluto/rea/pull/3456 **Author:** [@HawkForge](https://github.com/HawkForge) **Created:** 10/14/2026 **Status:** ✅ Merged **Merged:** 10/20/2026 **Merged by:** [@gensokyo-dev](https://github.com/gensokyo-dev) **Base:** `main` ← **Head:** `fix/timer-nonfinite-delay` --- ### 📝 Commits (3) - [`e76985f`](https://github.com/morluto/rea/commit/e76985fd938978822d18c782a286a1a205cac918) fix(javascript): keep overflowing timer delays unknown - [`30936f7`](https://github.com/morluto/rea/commit/30936f76b54902a280d1569b93028a78a34516e4) test(javascript): keep the IR-level overflowing timer assertion - [`e487a52`](https://github.com/morluto/rea/commit/e487a52ac86a4b3fb3f8882f619be88017cf6765) fix(javascript): retain JSON-safe timer facts through CLI and MCP ### 📊 Changes **3 files changed** (+131 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `src/domain/javascript/javascriptSemanticAnalysis.runtime_boundaries.test.ts` (+17 -0) 📝 `src/domain/javascript/javascriptSemanticAsyncEffects.ts` (+5 -1) ➕ `tests/boundary/mcp/javascriptTimerDelay.test.ts` (+109 -0) </details> ### 📄 Description ## Summary Fixes #3400. A numeric literal such as `1e999` becomes `Infinity` in the parsed AST. The timer fact producer passed it into JSON-backed properties; projecting one timer then failed the file's analysis and discarded unrelated function facts. The existing `timerDelay` owner now emits the established unknown value (`null`) for a non-finite literal. It preserves the timer, method, binding resolution, source range and file digest. Finite values, including `1e308`, remain unchanged. CLI and MCP consume this same representation; no downstream sanitization, schema relaxation or cache is added. A public CLI/stdio-MCP regression covers six timer cases, exact unknown-vs-omitted semantics, an unrelated exported function, source SHA-256, unchanged input and equal Evidence/node identities across adapters. The redundant new graph-only test is replaced by that stronger public workflow; the domain IR test retains its distinct binding-resolution assertion. Existing graph-consumer tests still run. ## Executed validation Pinned baseline **eac296a7d**, Linux x64, Node **24.18.0**, npm **11.16.0**: - **111 targeted tests passed**, covering runtime boundaries, dataflow, graph resources and the real CLI/MCP worker path. Full `npm run check` passed; `docs:check` passed, including **5 consistency tests**. - Integration with newer main **f5d1012fa**, tree `b219b2d6a6d3e8c3adec070ff81028ba1da97a92`: the public CLI/MCP regression passes again, and the real-package workflow retains the same 176 nodes / 4 selected rows with verified cleanup (**21.868 s** elapsed, **19.831 s** CPU, **1,077.8 MiB** peak). Main subsequently added only an unrelated Kuna test (`7b2087e44`), leaving production code unchanged. - The new public regression fails on exact main at the CLI success assertion. Independent real CLI runs reproduce `unreadable_output` / `JSON numbers must be finite` for both overflowing `setTimeout` and `setInterval`: main retains one module node and no timer or `greet`; candidate returns **14 nodes**, the unknown delay and `greet` for each unchanged source. - Six real CLI cases on each revision cover `1e999`, `1e309`, `100`, `1e308`, the `Infinity` identifier and a standalone overflowing literal. Healthy controls retain identical Evidence IDs between main and candidate. Source bytes remain unchanged; all owned temporary roots are empty. - Actual public MCP SDK **2026-07-28** on unmodified installed **fast-deep-equal 3.1.3**, selected `index.js` (**1,177 bytes**, SHA-256 `eb469e206280321a3878f2335ec98aa2104a155079d8ed83a23029098dccd215`): two summary analyses, a batch of **4 selected observation rows**, then full Evidence export. The retained graph has **176 semantic nodes**, including source function `equal`, with stable parent Evidence identity. Coverage remains **partial**: eight unselected files and 137 semantic unknowns are explicit. This is a small real package, not a complex-application performance claim. REA uses static source analysis; selected target files remain unchanged. The issue's original macOS/Node 26 environment was not separately exercised. ## Performance and context One CPU, 8-GiB hard / 7-GiB early process-family memory bounds, no swap. Equivalent completed package workflows include catalog discovery, two summaries, selected facts, Evidence export and shutdown: | Revision | Total elapsed | Family CPU | Peak family memory | First / repeat analysis | | --- | ---: | ---: | ---: | ---: | | Main | 19.856 s | 18.743 s | 1,075.4 MiB | 9.429 / 1.090 s | | Candidate | 20.967 s | 18.640 s | 1,147.9 MiB | 10.341 / 1.110 s | The repeat uses the same MCP session and immutable input; internal reuse is uninstrumented. No new cache is introduced. OS caches and shared-host load are uncontrolled; these samples establish no general speedup or precise latency effect. All completed tasks confirm cleanup and no OOM. The six-case CLI qualification bundles take **19.859 / 24.020 s**, CPU **18.237 / 21.525 s**, peak **384.5 / 384.3 MiB** (main/candidate). They intentionally differ in recovered facts, so their totals are not equivalent successful-workflow speed comparisons. `cl100k_base` estimates: compact input catalog **44,607 / 44,607 tokens**, selected analysis input **245 / 245**; the two summaries cost **2,338 tokens each**, selected facts **3,647**, and complete captured post-connect JSON traffic **732,048 / 732,051**. The **148,881-token** complete Evidence file is retained locally rather than delivered inline. Captured protocol includes output schemas; these are protocol/proxy estimates, not actual model context/usage. Initialization outside capture hooks and downstream prompts are unmeasured. Source selection, summary delivery and requested property keys bound work and inline detail while preserving full Evidence access. ## CI/review assessment and authorized bypass Latest head `e487a52ac`: Codex review completes with **no findings**. No actionable latest-head finding is open. Greptile's new run is still in progress; its earlier 5/5 is not claimed as a review of this head. Fork workflow approvals were supplied, but current classification/readiness/title jobs remain queued without completed test results. The task owner explicitly authorized bypass once independently qualified. Merge uses completed full checks, 111 targeted tests, docs checks, exact-base CLI failures, successful CLI/MCP regression, real-package analysis and newer-main integration. No queued CI or pending Greptile result is claimed as passing; subsequent results remain subject to assessment. Post-merge CI assessment: 35 latest-head checks succeeded; shard 1 failed only the existing `protobufCapture.test.ts` malformed-gRPC-web public workflow at its 30-second test timeout. Exact filtered controls reproduce the same timeout on pre-change main `cf5194f68` and candidate `e487a52ac`; the protobuf workflow is outside this diff. Other shards were cancelled or queued. No full CI pass is claimed. Latest Greptile review is **4/5**. Its outside-diff wrapped-timer cancellation lookup finding targets line 295, unchanged by this PR and already introduced on main in `6183544e8b`. The only production diff here is the `timerDelay` finite-number guard. The separate main concern is acknowledged with source attribution; it is not a bug introduced by this PR and is left outside this fix as the task owner requested. Codex has no findings. ## Author readiness Human owner: @gensokyo-dev The task owner explicitly delegated validation, readiness declarations and merge to this agent. These declarations are supplied under that authorization; no separate new manual human review by the original author is asserted. - [x] I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work. - [x] I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps. - [x] I declare this PR ready for initial maintainer review. ## Before merge - [x] Shared producer semantics and caller identity preserved; no new authority or schema changes. - [x] Relevant tests, real public workflows, performance and context costs verified; no local reports, binaries, credentials or dependencies in the diff. - [x] Available latest-head CI/review state and actionable findings assessed; queued work and bypass authorization recorded above. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-31 01:09:08 +00:00

closed this issue
added the
pull-request
label

bsakaguchi referenced this issue

2026-11-01 21:45:04 +00:00

[PR #5789] [MERGED] fix(model): preserve unsupported GGUF version diagnostics #7063

No milestone

No project

No assignees

1 participant

Notifications

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

morluto/rea#5749

No description provided.

Rows
Columns