deb

libnetplus (20260504+8) unstable; urgency=medium

  * QUIC: fix PTO re-queue spin loop — when sendPacket() failed (EAGAIN),

    the re-queued entry kept its original sent_time, causing

    checkLossAndRetransmit to fire PTO again immediately on the next call.

    This created an infinite hot loop that blocked the event loop.  Fix:

    reset sent_time to now and increment _pto_count for exponential

    back-off on send failure.

libnetplus (20260504+7) unstable; urgency=medium

  * QUIC: cap max_udp_payload_size to 1472 (Ethernet-safe) instead of

                    ++_pto_count;

                } else {

                    // Send failed (EAGAIN) — re-queue under original PN

                    // so the data is not permanently lost

                    // so the data is not permanently lost.  Reset sent_time

                    // to now and increment _pto_count so the next PTO check

                    // doesn't fire immediately (avoids hot spin-loop).

                    sp.sent_time = now;

                    ++_pto_count;

                    _sent_packets[old_pn] = std::move(sp);

                    std::cerr << "[QUIC-DIAG] PTO send failed, re-queued pn="

                              << old_pn << "\n";

                }

                QUIC_DBG("PTO probe: new_pn=%lu age=%.3fs pto=%.3fs",

                         (unsigned long)new_pn, age, pto_thresh);