AW-1: request() wraps a total ClientTimeout's bare asyncio.TimeoutError before
request_with_retries sees it, so the dedicated 'timeout' branch was dead and its comment
lied. wrap it as aiohttp.ServerTimeoutError (which IS both a ClientError AND a
TimeoutError) so direct request() callers still get a typed failure (M1 preserved) while
request_with_retries catches the timeout case first and labels it 'timeout'.
verified by execution: request() raises ServerTimeoutError (typed, M1 intact);
request_with_retries returns reason='timeout'; control confirms a real client error still
labels 'client error'. sibling-grep: aioweb_tls/aiowebhooks catch ClientError/TimeoutError,
both of which ServerTimeoutError satisfies — no consumer break.
Signed-off-by: disqualifier <dev@disqualifier.me>
- request_with_retries labels an exhausted total-timeout as 'timeout' instead of the
generic 'unexpected error' catch-all (nit)
- as_curl() renders an empty-but-valid json body ({} / []) via is-not-None instead of
dropping it as falsy (nit)
- _apply_overwrites snapshots the shared override dicts before iterating, so a
concurrent mutation can't raise 'dict changed size during iteration' (nit).
Signed-off-by: disqualifier <dev@disqualifier.me>
broaden the except to (aiohttp.ClientError, asyncio.TimeoutError) and re-wrap into
the same typed aiohttp.ClientError path. a total ClientTimeout raises a bare
asyncio.TimeoutError, which is NOT an aiohttp.ClientError subclass, so it previously
leaked raw out of request()/test_proxies(). add the missing asyncio import.
Signed-off-by: disqualifier <dev@disqualifier.me>
bump the commons dependency pin to v0.2.1 (retry attempts-floor fix). no code change;
the aretry migration is unaffected.
verified: 18/18 migration harness passes against commons 0.2.1.
Signed-off-by: disqualifier <dev@disqualifier.me>
- #7: request_with_retries routed only dicts to json=, so a valid JSON list body
was form-encoded via data=. add _route_body so dict OR list -> json=.
- #6: when every attempt returned a retryable status, the loop discarded the real
response and returned a synthetic FailureResponse (status 0). now the real last
4xx/5xx Response is returned on exhaustion (only a pure-exception failure yields
FailureResponse).
- migrate the retry/backoff loop onto commons.aretry (>=0.2.0); backoff schedule
unchanged (1,2,... = backoff_base**n), jitter off to match prior behavior.
verified by execution: list->json routing, exhausted 503 returns real 503 + body
with correct backoff, success/404 immediate, exception->falsy FailureResponse.
Signed-off-by: disqualifier <dev@disqualifier.me>
