fix: AnswerBuilder returned referenced documents in scrambled order

[Otis0408]

What's the problem

AnswerBuilder with return_only_referenced_documents=True collects the referenced document indices into a set and iterates it directly to build Answer.documents. Because CPython sets store items in internal hash-table order, the referenced documents were emitted in that hash-table order rather than ascending source-index order.

For small non-contiguous int indices this order is deterministic across runs (int hashing is identity), but it does not match the intuitive source order a user expects. For example, citations [3] [10] [50] yielded documents ordered 10, 3, 50.

Anyone rendering citations, feeding the referenced documents to a downstream component, or asserting on their order hits this: the documents look shuffled relative to how they were cited.

The fix

Iterate the collected indices via sorted(doc_idxs) so the referenced documents come out in ascending source-index order.

- for idx in doc_idxs:
+ for idx in sorted(doc_idxs):

Before / after (public component API)

from haystack import Document
from haystack.components.builders import AnswerBuilder

builder = AnswerBuilder(reference_pattern=r"\[(\d+)\]", return_only_referenced_documents=True)

result = builder.run(
    query="test query",
    replies=["First [3], then [10], and finally [50]."],
    documents=[Document(content=f"doc{i}") for i in range(1, 51)],
)

order = [doc.meta["source_index"] for doc in result["answers"][0].documents]
print(order)
# before: [10, 3, 50]   # set hash-table order (deterministic, but not source order)
# after:  [3, 10, 50]    # ascending source-index order

Notes

Wording-only, behavior-preserving fix plus one regression test. No public API change.

[vercel]

@Otis0408 is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
haystack/components/builders
answer_builder.py
Project Total

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