Closed huitema closed 11 months ago
In my tests of PR #1545, I saw that issue happen briefly. There were two reasons:
picoquic_retransmit_needed
that caused it to return "empty" packet, with length set to the header length, when there was no packet ready to be repeated.picoquic_prepare_packet-ready
.The header-only packet bug from loss recovery is being fixed in PR #1545. We need to add a unit test of ACK generation to ensure that if something else causes a similar issue, only a small number of ACK will be generated.
Perhaps related, but work on PR #1545 also discovered "too many ACK sent when using simple multipath", which could probably fixed at the same time as this "ACK loop" problem.
Reported as not reproducing after PR #1545.
Added tests and coding safeties in Pr #1568.
Client (on a MAC) is connected to a local server using loopback. The client drops abruptly. We would expect to see a few retransmission attempts before the server gives up on the connection, but instead we see the servers sending ACK messages in a loop.