fix: deep-copy document metadata in PythonCodeSplitter

[sainikhiljuluri]

Problem

PythonCodeSplitter copies the input document's metadata shallowly, so nested values are shared between every chunk and with the caller's document:

doc = Document(content=class_source, meta={"tags": ["code"]})
chunks = PythonCodeSplitter(min_effective_lines=2, max_effective_lines=5).run(documents=[doc])["documents"]

chunks[0].meta["tags"] is chunks[1].meta["tags"]   # True
chunks[0].meta["tags"] is doc.meta["tags"]         # True

chunks[0].meta["tags"].append("shape")
chunks[1].meta["tags"]   # ['code', 'shape']
doc.meta["tags"]         # ['code', 'shape']   <- the caller's document is mutated

Root cause

Two shallow copies:

• _build_chunk_meta does meta.update({k: v for k, v in parent_doc.meta.items() ...}), so every chunk holds references into the input document's metadata.
• _secondary_split does meta = dict(base_meta), so the pieces of an oversized unit share the same nested objects with each other.

Fix

Deep-copy at both sites, matching the other splitters.

This is the same defect as #12248, fixed for the Markdown, CSV and hierarchical splitters by #12249. PythonCodeSplitter was not covered there. I checked the remaining splitters — DocumentSplitter, RecursiveDocumentSplitter, CSVDocumentCleaner and DocumentCleaner are already safe, so this should close out the family.

Tests

Two regression tests in TestFileNamePropagation, one per fixed site, written in the same shape as the one #12249 added to test_markdown_header_splitter.py.

Reverting the fix fails both, on the leaked mutation itself:

assert ['code', 'shape'] == ['code']
assert ['code', 'giant'] == ['code']

test/components/preprocessors/ is green: 369 passed, 10 skipped (69 -> 71 in this file). ruff check and ruff format --check clean.

Risk

Behavioural change is limited to chunks no longer aliasing the caller's metadata. Deep-copying runs once per chunk on metadata that was already being copied, so the added cost is proportional to metadata size, not document size.

---

AI assistance was used to write this change.

[vercel]

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

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[HaystackBot]

Hi @sainikhiljuluri, thanks a lot for your contribution! 🙏

We noticed that the Contributor License Agreement (CLA) check (license/cla) hasn't passed yet, so we've temporarily moved this PR to draft and paused the review assignment.

To get your PR reviewed, please sign the CLA via the link in the license/cla check below (or in the CLA bot comment). As soon as the check turns green, this PR will automatically be marked ready for review again and a reviewer will be re-assigned.

[HaystackBot]

Thanks for signing the CLA, @sainikhiljuluri! 🎉 This PR is now ready for review again and the reviewer has been re-assigned.