test: enable mypy typing checks for test/components/preprocessors

[ShousenZHANG]

Related Issues

• part of Run typing checks on our test files #10396

Claimed in this comment.

Proposed Changes:

test/components/preprocessors/ was not in the repository mypy target, so its 13 modules were not checked by hatch run test:types. This adds the path and fixes the errors that surfaced.

The baseline was 265 errors, but 206 of them were in test_recursive_splitter.py alone, cascading from one pattern repeated seven times where the dict returned by run() was rebound to the list it contains:

doc_chunks = splitter.run([doc])
doc_chunks = doc_chunks["documents"]   # dict -> list, same name

Naming the two values separately drops that file from 206 errors to 26, and the directory from 265 to 85.

The remaining 85 are mostly Document.content being str | None. AGENTS.md asks to avoid type: ignore, casts and assertions where possible and to explain them where they are necessary, so:

• Narrowing assert ... is not None before first use, where the value was already being dereferenced. cast(str, ...) would type-check equally well but keeps the old behaviour of failing with a TypeError from inside a comprehension; the assert names the expectation at the point it is made. cast also appears only twice in all of test/.
• Comprehensions and generators over .content expanded into loops. A narrowing in an enclosing block does not reach a comprehension's own scope, even when the comprehension reuses the same name, so the assert has to live inside the iteration. The alternative already merged in this campaign is test: enable mypy typing checks for test/components/samplers/ #12433's in-comprehension filter plus a compensating assert len(filtered) == len(docs). That is two lines instead of four, but it is only equivalent as long as every one of the 14 sites remembers the length guard; dropping it silently skips elements. I went with the loop because it cannot be got wrong site-by-site — happy to convert to the filter idiom if you prefer the smaller diff.
• any(...) kept as any(...) over the collected list rather than expanded into per-element asserts, since turning an existential check into a universal one would change what the test claims.

Two smaller groups:

• Six # type: ignore comments on read_csv(...) in test_csv_document_splitter.py are stale. warn_unused_ignores flags them as soon as the directory enters the target, so removing them is required rather than optional.
• Eight tests pass a wrong type on purpose to assert the raised error. Those keep the value and carry a specific ignore code ([arg-type] / [list-item]), never a bare one.

test_document_preprocessor.py needed a different fix: Pipeline.get_component() returns the Component protocol, so the attribute reads failed. An isinstance narrowing resolves it, the same way #12343 and #12319 did.

No test assertion, expected value, or input datum changed. Assert counts per file only go up or stay flat.

How did you test it?

• hatch run test:types -> Success: no issues found in 481 source files (468 before).
• hatch run test:unit test/components/preprocessors/ -> 361 passed, 19 deselected — identical to the counts on main before the change.
• hatch run test:integration test/components/preprocessors/ -> 9 passed, 10 skipped. The 10 skipped are the OPENAI_API_KEY-gated embedding-splitter tests; six of the edits land in those bodies, so they are covered by mypy and collection but were not executed. To check those six anyway I ran their pre- and post-change versions side by side against a stubbed embedder over a range of mutated outputs, and the two behave identically.
• hatch run fmt-check -> clean across the repo (6787 files).

Notes for the reviewer

This and my #12481 both add a directory to the same types = line, so whichever lands second will conflict there and I will rebase it — no need to sequence them on my account.

Nine open PRs add tests to these files (#12284, #12346, #12356, #12373, #12424, #12431, #12434, #12468, #12478). All are pure additions and a three-way merge against each is clean in both orders, so this does not block them. Two of them (#12284, #12478) add a Document.content dereference that mypy will start checking once this lands; I will leave them a note so their CI does not go red unexpectedly.

No release note: CONTRIBUTING.md scopes the requirement to user-facing changes and notes that tests-only changes can carry ignore-for-release-notes — happy for you to add the label if you would rather have it explicit.

This PR was fully generated with an AI assistant. I have reviewed the changes and run the relevant tests.

Checklist

• I have read the contributors guidelines and the code of conduct.
• I have updated the related issue with new insights and changes.
• I have added unit tests and updated the docstrings - n/a, test-only typing change with no behavior or public API change.
• I've used one of the conventional commit types for my PR title.
• I have documented my code.
• I have added a release note file - not needed here, see above.
• I have run pre-commit hooks and fixed any issue.

[vercel]

@ShousenZHANG is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
haystack/core/pipeline
pipeline.py
Project Total

This report was generated by python-coverage-comment-action

[sjrl]

Thanks!