[GH-ISSUE #5802] perf(rust): query selected dependency facts from retained Evidence #3513

New issue

Closed

opened 2026-10-31 00:41:20 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 00:41:20 +00:00

Owner

Originally created by @gensokyo-dev on GitHub (Oct 21, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/5802

Problem and observations

At main d672cc23998b, inspect_rust_binary accepts only path and returns the complete embedded dependency graph. A question about exact crate versions and a few dependency edges requires receiving all package/edge rows. Unlike existing Go/model workflows, Rust has no summary or retained selected-query path. Provider admission also serializes the entire report just to count bytes; later service validation and repeated source inspections add work.

Executed baseline, pinned Alpine v3.21 x86_64 packages:

ripgrep14.1.1-r0:42 packages /88 edges; complete CLI reply10,605 bytes.
Zellij0.40.0-r0:422 packages /1,063 edges; complete CLI reply86,805 bytes.
GNU objcopy section extraction plus independent Python zlib/JSON matches every normalized package and edge, source section bytes and full artifact SHA-256. Applications were not executed. One-CPU4-GiB hard /3-GiB early-stop public CLI controls qualify with verified exit/cleanup and no limit events. These are observed reply sizes/correctness, not token or speedup claims. The completed equivalent SDK cohorts, token estimates, negative timing results and controls are now recorded in PR #5815 and experiments/analysis-efficiency/rust-dependency-views.md.

Intended task and scope

Answer an exact embedded crate/version/source question, retrieve the associated direct dependency relationships in batches, and export complete Evidence. Keep original package/edge identity and all matching versions, source provenance, null/missing-section distinctions, cycles, inferred compiler facts and uncertainty about actual runtime dependencies.

Extend the existing inspect and inspect_analysis_view workflows, not a new tool:

opt-in summary/direct selected delivery while preserving omitted/explicit complete behavior;
original-index package facts and edge pages, exact batch filters, selected fields and matching-row paging;
shared domain/application projection for CLI and MCP, retained reads without reopening the source;
reuse validation only for authenticated immutable Evidence within its lifetime; no live-file result cache;
evaluate temporary JSON sizing copies; the prototype replacement was reverted after inconsistent results, leaving producer admission unchanged.

This remains existing read-only ELF/PE/thin-Mach-O scope. No target execution, injection, mutation, installer/dependency change or new maintainer authority is involved. Avoid overlap with Rust WASM support #5069 and existing correctness fixes.

Acceptance / verification

Independent real-artifact facts and complete original Evidence identities agree before/after, including after removing an owned input copy.
Exact selection intersects filters; duplicates/variants, unknowns, missing metadata, explicit detail and source/subject mismatch remain meaningful.
Real CLI and production stdio SDK verify converted named schemas, selected answers, retained export and cleanup. Keep malformed, capacity, cancellation and cleanup regressions.
Compare equivalent completed tasks and a small control; include catalog/schemas, extra calls/retries, first/repeat timings, CPU and peak process-family memory. Label token estimates versus actual model usage and preserve negative results.
Refresh main before integration, use this issue's branch, run relevant checks, ready PR, automated review/CI and merge. No claim that every Rust workflow is qualified.

Related: coverage continuation #3135 and original Rust feature #2456.

Originally created by @gensokyo-dev on GitHub (Oct 21, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/5802 ## Problem and observations At main `d672cc23998b`, `inspect_rust_binary` accepts only `path` and returns the complete embedded dependency graph. A question about exact crate versions and a few dependency edges requires receiving all package/edge rows. Unlike existing Go/model workflows, Rust has no summary or retained selected-query path. Provider admission also serializes the entire report just to count bytes; later service validation and repeated source inspections add work. Executed baseline, pinned Alpine v3.21 x86_64 packages: - ripgrep14.1.1-r0:42 packages /88 edges; complete CLI reply10,605 bytes. - Zellij0.40.0-r0:422 packages /1,063 edges; complete CLI reply86,805 bytes. GNU objcopy section extraction plus independent Python zlib/JSON matches every normalized package and edge, source section bytes and full artifact SHA-256. Applications were not executed. One-CPU4-GiB hard /3-GiB early-stop public CLI controls qualify with verified exit/cleanup and no limit events. These are observed reply sizes/correctness, not token or speedup claims. The completed equivalent SDK cohorts, token estimates, negative timing results and controls are now recorded in PR #5815 and experiments/analysis-efficiency/rust-dependency-views.md. ## Intended task and scope Answer an exact embedded crate/version/source question, retrieve the associated direct dependency relationships in batches, and export complete Evidence. Keep original package/edge identity and all matching versions, source provenance, null/missing-section distinctions, cycles, inferred compiler facts and uncertainty about actual runtime dependencies. Extend the existing inspect and inspect_analysis_view workflows, not a new tool: - opt-in summary/direct selected delivery while preserving omitted/explicit complete behavior; - original-index package facts and edge pages, exact batch filters, selected fields and matching-row paging; - shared domain/application projection for CLI and MCP, retained reads without reopening the source; - reuse validation only for authenticated immutable Evidence within its lifetime; no live-file result cache; - evaluate temporary JSON sizing copies; the prototype replacement was reverted after inconsistent results, leaving producer admission unchanged. This remains existing read-only ELF/PE/thin-Mach-O scope. No target execution, injection, mutation, installer/dependency change or new maintainer authority is involved. Avoid overlap with Rust WASM support #5069 and existing correctness fixes. ## Acceptance / verification - Independent real-artifact facts and complete original Evidence identities agree before/after, including after removing an owned input copy. - Exact selection intersects filters; duplicates/variants, unknowns, missing metadata, explicit detail and source/subject mismatch remain meaningful. - Real CLI and production stdio SDK verify converted named schemas, selected answers, retained export and cleanup. Keep malformed, capacity, cancellation and cleanup regressions. - Compare equivalent completed tasks and a small control; include catalog/schemas, extra calls/retries, first/repeat timings, CPU and peak process-family memory. Label token estimates versus actual model usage and preserve negative results. - Refresh main before integration, use this issue's branch, run relevant checks, ready PR, automated review/CI and merge. No claim that every Rust workflow is qualified. Related: coverage continuation #3135 and original Rust feature #2456.

bsakaguchi closed this issue

2026-10-31 00:41:20 +00:00

bsakaguchi referenced this issue

2026-10-31 01:09:28 +00:00

[PR #3513] fix(browser): keep an undecodable Electron worker target uncertain #5792

bsakaguchi referenced this issue

2026-11-01 21:11:07 +00:00

[GH-ISSUE #5654] [Tracker] Open PRs that can't merge yet: conflicts and red CI (October 2026) #3425