[Sanjays2402]

Summary

The warning logged when get_document_from_dict finds a document without text was showing data.keys() — but the default adapter already popped keys into parsed_data["text"] and parsed_data["metadata"], so by the time the warning fires the raw dict is empty. A user sees Available keys: [] and has no idea if their text_key is wrong or the field just has an empty value.

Fix

Snapshot original_keys before the adapter runs, then:

• If text_key was not found — show the real available keys
• If text_key was found but the value is empty — say so explicitly

Closes #341.

---

Note

Low Risk
Logging-only change in the reader skip path; document filtering behavior is unchanged aside from clearer diagnostics.

Overview
Fixes misleading “Available keys: []” logs when BaseReader.get_document_from_dict skips a document with no text.

The default adapter mutates the raw record (pops text_key, metadata, etc.) before the warning runs, so the old message used post-adapter data.keys(). The change captures original_keys before calling the adapter, then branches the one-time warning: if text_key was missing, it lists those keys; if text_key was present but empty, it says the value was empty instead of implying a wrong key name.

^{Reviewed by Cursor Bugbot for commit 0eb3dd8. Bugbot is set up for automated code reviews on this repo. Configure here.}