Dropped extracted decisions and preferences still write their steps and entities

オープン
#342 コメント 2 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
3/5
見積もり時間
1〜2日
初心者へのやさしさ
66/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
python, typescript
領域
backend, databases

調査の方向性

Start with _assemble in src/selfgraph/extraction/extractor.py, focusing on the decision and preference loops and how add_relation and resolve() register writes. Reproduce both cases from the issue against embedded Kuzu, then inspect the relevant extraction tests. Done when dropped items leave no relations or fallback entities in the graph, while valid items still persist.

索引モデルが issue の本文から書いたものです。

説明

approved-for-pr area: extraction bug effort: S P2

What happens

When insert.text drops an extracted decision or preference as invalid, parts of it are still written:

  • A decision dropped for an empty or control-character title still writes the relations its steps named.
  • A preference dropped for a bad strength or key still creates its subject and value as Thing entities.

The warning says the item was dropped, but the graph holds what the item created.

Reproduce

On main (5a277c9), embedded Kuzu:

import asyncio, tempfile
from selfgraph import KuzuStore, SelfGraph

class LLM:
    def __init__(self, **replies):
        self.replies = replies

    async def generate(self, *, system, user, output):
        kind = next(iter(output.model_fields))
        return {kind: self.replies.get(kind, [])}

async def run(llm):
    with tempfile.TemporaryDirectory() as tmp:
        async with SelfGraph(store=KuzuStore(f"{tmp}/g.kuzu"), llm=llm) as g:
            r = await g.insert.text("x")
            print([w.message for w in r.warnings])
            print([e.id for e in r.entities], [(x.from_entity, x.type, x.to_entity) for x in r.relations])

asyncio.run(run(LLM(
    entities=[{"type": "Person", "name": "Ann"}, {"type": "Company", "name": "Acme"}],
    decisions=[{"title": "", "actor_name": "Ann", "steps": [{"from_name": "Ann", "type": "CHOSE", "to_name": "Acme"}]}],
)))
asyncio.run(run(LLM(
    preferences=[{"subject_name": "Zed", "key": "food", "value_name": "Pizza", "strength": 5.0}],
)))

The first run warns dropped an extracted decision: decision title cannot be empty and stores person:ann -CHOSE-> company:acme. The second warns about the strength and stores thing:zed and thing:pizza.

Cause

src/selfgraph/extraction/extractor.py, _assemble: the decision loop calls add_relation for each step, and the preference loop calls resolve() (which registers fallback entities), before the item's own checks run. Validating the item first, or registering its writes only once it is accepted, would fix both.

主要言語
TypeScript
スター
173
フォーク
58
平均マージ
6時間
マージ済み PR(30日)
120

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

hydra-db/selfgraph のほかの issue

hydra-db/selfgraph の issue をすべて見る

似ている issue

TypeScript の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。