[lianbie]

Summary

Bring api/tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py under strict Pyrefly checking and remove its entry from api/tests/unit_tests/pyrefly.toml, continuing #24645 (same pattern as #43883, #43875 and #43798).

The module was still excluded from strict checking. Strict Pyrefly (1.4.0) reports 19 diagnostics on it, all fixed here:

Diagnostic	Count	Fix
unannotated-return	10	-> None on the eight tests, _DummyTFIDF.__init__ and _install_fake_jieba_modules
implicit-any-parameter	2	monkeypatch: pytest.MonkeyPatch, **kwargs: object
missing-attribute	7	document the intentional fake-module attribute assignments with # pyrefly: ignore[missing-attribute]

The last group is by design: the tests install a fake jieba.analyse with types.ModuleType, then supply default_tfidf, TFIDF and analyse for the handler to discover. A ModuleType declares none of those, so the assignment is correct at runtime and the suppression states why. typing.Any is not used, per the banned-API rule.

No assertion, fixture, import list (apart from the dropped Any) or test behavior changes.

What changed

• api/tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py: annotations and the seven scoped suppressions above.
• api/tests/unit_tests/pyrefly.toml: remove the single exclusion line.

Validation

• pyrefly check --config=tests/unit_tests/pyrefly.toml --use-ignore-files=false --disable-project-excludes-heuristics=true tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py: 19 diagnostics before, 0 after (pyrefly 1.4.0).
• The eight tests in the module pass with the edited file.
• ruff check and ruff format --check pass on the changed file.
• No production code is touched, so no runtime behavior can change.

Contributes to #24645

[github-actions]

Pyrefly Type Coverage

Metric	Base	PR	Delta
Type coverage	83.19%	83.20%	+0.01%
Strict coverage	82.84%	82.85%	+0.01%
Typed symbols	85,935	85,943	+8
Untyped symbols	17,438	17,429	-9
Modules	4378	4378	0

[github-actions]

Pyrefly Diff

base → PR

--- /tmp/pyrefly_base.txt	2026-11-04 07:10:21.954836148 +0000
+++ /tmp/pyrefly_pr.txt	2026-11-04 07:10:10.482801959 +0000
@@ -4338,20 +4338,6 @@
   --> tests/unit_tests/core/rag/data_post_processor/test_data_post_processor.py:29:29
 ERROR Argument `dict[str, str]` is not assignable to parameter `reranking_model` with type `RerankingModelDict | None` in function `core.rag.data_post_processor.data_post_processor.DataPostProcessor._get_rerank_model_instance` [bad-argument-type]
    --> tests/unit_tests/core/rag/data_post_processor/test_data_post_processor.py:192:33
-ERROR Object of class `ModuleType` has no attribute `analyse` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:32:5
-ERROR Object of class `ModuleType` has no attribute `default_tfidf` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:46:5
-ERROR Object of class `ModuleType` has no attribute `default_tfidf` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:58:5
-ERROR Object of class `ModuleType` has no attribute `TFIDF` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:63:5
-ERROR Object of class `ModuleType` has no attribute `default_tfidf` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:74:5
-ERROR Object of class `ModuleType` has no attribute `TFIDF` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:81:5
-ERROR Object of class `ModuleType` has no attribute `default_tfidf` [missing-attribute]
-  --> tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py:92:5
 ERROR Argument `str` is not assignable to parameter `doc_form` with type `IndexStructureType | SQLCoreOperations[IndexStructureType]` in function `models.dataset.Document.__init__` [bad-argument-type]
    --> tests/unit_tests/core/rag/datasource/test_retrieval_service.py:148:18
 ERROR Argument `list[FromClause]` is not assignable to parameter `tables` with type `Sequence[Table] | None` in function `sqlalchemy.sql.schema.MetaData.create_all` [bad-argument-type]

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.84%. Comparing base (d1e859d) to head (14706e0).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #45320      +/-   ##
==========================================
- Coverage   89.84%   89.84%   -0.01%     
==========================================
  Files        6242     6242              
  Lines      391154   391154              
  Branches    79217    79217              
==========================================
- Hits       351445   351432      -13     
- Misses      33607    33618      +11     
- Partials     6102     6104       +2     

Flag	Coverage Δ
api	89.64% <ø> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[lianbie]

@asukaminato0721 I addressed the two things you asked for, in 20a2767:

• Replaced _DummyTFIDF with _RecordingTFIDF(TFIDF), a subclass of the real jieba.analyse.TFIDF. The real TFIDF.__init__ runs, so the IDF table is built and stop_words exists; only extract_tags returns a test-chosen keyword list.
• The loader branches now assert against the real class (isinstance(handler._tfidf, TFIDF)), instead of the two ad-hoc subclasses the previous revision used.

On the new head: strict pyrefly is 0 diagnostics for this file, the 6 tests pass, and ruff check / ruff format are clean.

@GYJ99 thanks for flagging the overlap with #45347 on that other PR — agreed this should land as one change, and #45320 already uses the real TFIDF rather than Any.

Also noting for maintainers: every workflow on 20a2767 is action_required, and I cannot approve runs for the upstream repo from my fork account — CI will need one maintainer click to get going. (The previous revision, e703597, had all workflows green.)

[lianbie]

Friendly follow-up on review: the requested change is in and the branch is now clean.

Addressed: _DummyTFIDF is replaced by _RecordingTFIDF(TFIDF), a subclass of the real jieba.analyse.TFIDF — the production __init__ runs, so the IDF table is built and stop_words exists; only extract_tags returns a test-chosen keyword list. The loader branches assert with isinstance(handler._tfidf, TFIDF) against the real class. Replied in the inline thread as well.

Formatting: autofix picked up my file in d0e9b6a and moved from jieba.analyse import TFIDF into the third-party block, so ruff check / format / import ordering now match the repo config.

One note on the Pyrefly report, in case it is read as a regression: Type coverage 83.11% -> 82.89% is not from this file. The diff shows seven missing-attribute diagnostics removed from test_jieba_keyword_table_handler.py (the file is also removed from tests/unit_tests/pyrefly.toml). The added diagnostics in the same report are in test_app_config_manager.py, test_generator.py, test_session_store.py and similar files that this PR does not touch; they come from base-branch drift between the two runs.

Full CI still shows action_required on every workflow for d0e9b6a — that needs a maintainer approval to run, which I cannot grant from my fork account. The previous revision (953de927) had all workflows green apart from autofix.ci, which is what produced d0e9b6a.

Could you take another look when you have a moment? Thanks.