Rust-hybrid wall-clock A/B v2: baseline profile and candidate selection

[jununfly]

This was generated by AI during triage.

Context

Parent plan: docs/plans/2026-06-19-rust-hybrid-default-indexing-wall-clock-ab.md

References:

• Post-PRD optimization tracker for Rust indexing performance #165 long-running Rust indexing performance tracker
• Profile Rust parseExtractionMs before next optimization #224 existing parse-extraction profiling candidate
• Prior bounded A/B pass: Rust-hybrid performance A/B: baseline profile and candidate selection #287-Rust-hybrid indexing performance A/B tracker #290

This issue starts the next rust-hybrid default-path full-index wall-clock A/B loop after the prior Rust write-path optimization. Do not change production code in this issue. The job is to collect a credible current baseline and select exactly one bounded candidate, or record no-go.

Scope

Use source-path rust-hybrid full indexing only.

Corpora:

• /Users/bilibili/Documents/workspace/jununfly/ZCodeGraph
• /private/tmp/codegraph-corpus/vscode-sparse

The VS Code sparse checkout must already exist and be a Git checkout. If it is missing or not a Git checkout, stop that portion and mark this issue as needing human setup. Do not clone or replace the corpus from the agent.

Acceptance Criteria

• Verify both corpus paths, including Git checkout status for VS Code sparse.
• Run baseline source-path rust-hybrid full-index profiles on the current repo and VS Code sparse when available.
• Record wall-clock time.
• Record RSS / memory stability, or a clear unavailable reason.
• Record diagnostic buckets for rustCore.parseExtractionMs, rustCore.sqliteWriteMs, typescriptFinalizationMs, finalize.referenceResolutionMs, public finalization sub-buckets, and fallback append/accounting where available.
• Decide whether Profile Rust parseExtractionMs before next optimization #224 is the next candidate, not selected, or needs reframing.
• Select exactly one bounded optimization candidate, or record no-go.
• Do not change production code.
• Do not run full benchmark scoreboard or agent sufficiency A/B.
• Save evidence under docs/benchmarks/.

[jununfly]

This was generated by AI during triage.

Baseline evidence is recorded in docs/benchmarks/2026-06-19-rust-hybrid-wall-clock-ab-v2-baseline.md, with raw profile artifacts for ZCodeGraph and VS Code sparse.

Decision: proceed to #292 with exactly one bounded candidate: reuse one tree-sitter parser per source language during Rust core extraction instead of constructing/configuring a new parser for every file.

Key evidence:

• VS Code sparse wall-clock: 295.14s; RSS: 2,321,973,248 bytes.
• VS Code sparse rustCore.parseExtractionMs: 40,052ms.
• VS Code sparse rustCore.sqliteWriteMs: 54,061ms.
• VS Code sparse typescriptFinalizationMs: 126,363ms.

Interpretation:

• TypeScript finalization is still the largest end-to-end bucket, but the low-semantic-risk options there are less isolated from this baseline and must not drift into reference-disambiguation semantics.
• Profile Rust parseExtractionMs before next optimization #224 remains relevant and is partially reframed: this pass tries one narrow parse/extraction overhead candidate before opening broader parser/extractor subsegment work.
• No production code changed in Rust-hybrid wall-clock A/B v2: baseline profile and candidate selection #291; no full scoreboard or agent sufficiency A/B was run.