arxiv.get_metadata doctest reports KeyError doi instead of the arXiv error #2805

✍️🤖 Agent/Nous: on behalf of @WolfgangFahl

In two CI runs on 2026-10-02 and 2026-10-03 (commit 1626b5e) the doctest of scholia.arxiv.get_metadata failed with

>>> metadata = get_metadata(1503.00759)
>>> metadata[doi] == 10.1109/JPROC.2015.2483592
UNEXPECTED EXCEPTION: KeyError(doi)

A rerun shortly afterwards passed. The arXiv API delivers arxiv:doi 10.1109/JPROC.2015.2483592 for 1503.00759 and feedparser 6.0.14 maps it to arxiv_doi, so the data and the parsing are fine.

The KeyError hides the real cause: on a non 200 status, a network error or any other exception get_metadata returns a dict with only the key error, so the doctest cannot show what arXiv answered. The broad except Exception also hides parsing errors.

Proposal:

• the doctest shows the error together with the doi, e.g.

>>> metadata = get_metadata(1503.00759)
>>> metadata.get(error), metadata.get(doi)
(None, 10.1109/JPROC.2015.2483592)

• the error message carries the HTTP status or the exception text
• the live arXiv doctest is skippable in the same way as the live Wikidata API doctests of Doctests in scholia/api.py fail while the Wikidata API reports maxlag #2804