Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.
Python aicode-complexitycode-healthcode-intelligencecode-qualitydead-codedeveloper-toolsdocumentationgit-analyticsmcpopen-sourcepythonrefactoringstatic-analysistechnical-debt
"""The vector arm's evidence snippet, and the field it is cut from.
The full-text arm holds the page content at search time and can window it. The
vector arm cannot: LanceDB stores a fixed prefix of the content at index time
and that is all a search has to work with. So the stored prefix has to be wide
enough for a window to be cut out of it.
Two other readers of the same column use it as a short prompt summary and must
keep the size they had — widening the store must not widen a prompt.
"""
from __future__ import annotations
import pytest
from repowise.core.persistence.search import _SNIPPET_LEN
from repowise.core.persistence.vector_store.lancedb_store import (
STORED_SNIPPET_CHARS,
LanceDBVectorStore,
)
from repowise.core.providers.embedding.base import MockEmbedder
_OPENER = "## Overview\n\nThis module is part of the indexing pipeline. " + "Filler. " * 60
_MATCH = "The walker refuses to descend into a nested git checkout."
@pytest.fixture
async def store(tmp_path):
st = LanceDBVectorStore(db_path=str(tmp_path / "lancedb"), embedder=MockEmbedder())
yield st
def _page(content: str) -> dict:
return {
"title": "File: walker.py",
"page_type": "file_page",
"target_path": "src/walker.py",
"content": content,
}
class TestStoredWidth:
def test_the_stored_prefix_is_wider_than_the_snippet_it_serves(self):
"""A window can only be cut from text the store actually kept."""
assert STORED_SNIPPET_CHARS > _SNIPPET_LEN
async def test_a_long_page_is_stored_up_to_the_cap(self, store):
content = _OPENER + _MATCH + " tail " * 1000
await store.embed_and_upsert("p1", content, _page(content))
rows = await store._table.query().select(["content_snippet"]).to_list()
assert len(rows[0]["content_snippet"]) == STORED_SNIPPET_CHARS
async def test_a_page_shorter_than_the_cap_is_stored_whole(self, store):
content = _OPENER + _MATCH
assert len(content) < STORED_SNIPPET_CHARS
await store.embed_and_upsert("p1", content, _page(content))
rows = await store._table.query().select(["content_snippet"]).to_list()
assert rows[0]["content_snippet"] == content
class TestSearchSnippet:
async def test_a_query_carrying_search_centres_the_snippet(self, store):
content = _OPENER + _MATCH
await store.embed_and_upsert("p1", content, _page(content))
results = await store.search("nested git checkout", limit=5)
assert "nested git checkout" in results[0].snippet
assert len(results[0].snippet) <= _SNIPPET_LEN
async def test_no_match_falls_back_to_the_opener(self, store):
content = _OPENER + _MATCH
await store.embed_and_upsert("p1", content, _page(content))
results = await store.search("xylophone tuning", limit=5)
assert results[0].snippet.startswith("## Overview")
async def test_searching_by_raw_vector_serves_the_opener_at_the_old_size(self, store):
"""That path has no query text, so there is nothing to centre on.
It must not start returning the whole stored prefix instead — this is
what the answer pipeline calls, and the snippet reaches a prompt.
"""
content = _OPENER + _MATCH
await store.embed_and_upsert("p1", content, _page(content))
vector = (await store._embedder.embed(["anything"]))[0]
results = await store.search_by_vector([float(v) for v in vector], limit=5)
assert len(results[0].snippet) <= _SNIPPET_LEN
assert results[0].snippet.startswith("## Overview")
class TestPromptSummariesKeepTheirSize:
"""The prompt-injection readers of this column must not grow with it."""
async def test_summary_by_path_stays_at_the_old_width(self, store):
content = _OPENER + _MATCH
await store.embed_and_upsert("p1", content, _page(content))
found = await store.get_page_summary_by_path("src/walker.py")
assert len(found["summary"]) <= _SNIPPET_LEN
async def test_summaries_by_paths_stay_at_the_old_width(self, store):
content = _OPENER + _MATCH
await store.embed_and_upsert("p1", content, _page(content))
found = await store.get_page_summaries_by_paths(["src/walker.py"])
assert len(found["src/walker.py"]["summary"]) <= _SNIPPET_LEN