When the wrapper invoked `python3 - <<'PY' ... PY` inside a function that
was being fed JSON via a pipe (`printf '%s' "$STATUS_JSON" |
extract_state_from_status_json`), the heredoc bound stdin to the Python
program text. The `-` argument tells Python to read its program from
stdin, so the program consumed stdin before json.load(sys.stdin) ran —
which then saw EOF and bailed to the "unknown" branch every time.
Result: pr-ci-wait.sh hung the full timeout (default 30 min) even when
the upstream Gitea status was already 'success', because every poll
returned 'unknown' and the loop kept retrying.
Fix: capture the piped JSON into a local variable with `payload=$(cat)`
BEFORE invoking python, then pass it via env (PR_CI_STATUS_JSON). The
heredoc still drives the Python program, but the payload is now read
from the environment instead of a stdin that's already been consumed.
Same fix applied to print_status_summary() which has the identical
pattern.
Verified locally:
$ echo '{"state":"success"}' | extract_state_from_status_json → success
$ echo '' | extract_state_from_status_json → unknown
$ echo '{"state":null,"statuses":[{"state":"success"}]}' | … → success
$ echo '{"state":"pending"}' | extract_state_from_status_json → pending
$ echo '{"state":"failure"}' | extract_state_from_status_json → failure
Reported in screenshot from operator session 2026-05-13 — wrapper was
stuck waiting on a PR whose underlying Gitea status was already
success. Operator workaround was to bypass the wrapper and use the raw
Gitea API.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
e2d49aface