fix: Trim two comments on the data source shutdown and retry wiring

Addresses review feedback on #519.

AsyncPollingUpdateProcessor.stop() now carries one line saying why it
waits before closing the transport. The previous wording read as if the
processor kept working after OFF, when it is shutting down.

The comment on the bare RetryDelayStrategy keeps only the fact a reader
needs: the strategy must be passed, or the SSE client picks its own
backoff.

Author: jsonbailey

ldclient/impl/aio/transport.py

@@ -141,8 +141,7 @@ def create(self, url: str, initial_retry_delay: float, query_params=None, sdk_ma
             aiohttp_request_options["proxy"] = proxy
         if sdk_managed_retry:
             # The SSE client's retry is disabled; the SDK owns the delay. The base
-            # strategy returns the delay unchanged, so the wait is always zero;
-            # omitting it would select the library's own backoff.
+            # strategy must be passed: omitting it selects the library's backoff.
             retry_options: dict = {
                 "initial_retry_delay": 0,
                 "retry_delay_strategy": RetryDelayStrategy(),

ldclient/impl/datasource/async_polling.py

@@ -63,10 +63,7 @@ async def stop(self):
         if self._data_source_update_sink is not None:
             self._data_source_update_sink.update_status(DataSourceState.OFF, None)
 
-        # OFF is reported first, so a listener sees the shutdown at once. The wait
-        # that follows only drains a poll already in flight, so the transport is
-        # not closed while that request still uses it. The close is in a finally,
-        # so an owned transport is released even if stop() is cancelled mid-wait.
+        # Do not close the transport while an in-flight request still uses it.
         try:
             await self._task.wait_stopped()
         finally:

ldclient/impl/datasource/streaming.py

@@ -141,8 +141,7 @@ def _create_sse_client(self) -> SSEClient:
             ),
             error_strategy=ErrorStrategy.always_continue(),  # we'll make error-handling decisions when we see a Fault
             # The SSE client's retry is disabled; the SDK owns the delay. The base
-            # strategy returns the delay unchanged, so the wait is always zero;
-            # omitting it would select the library's own backoff.
+            # strategy must be passed: omitting it selects the library's backoff.
             initial_retry_delay=0,
             retry_delay_strategy=RetryDelayStrategy(),
             retry_delay_reset_threshold=0,