Repository navigation
Description
Symptom
After JSR net_tcp_connect, callers observe net_tcp_state = UCI_TCP_CONNECTED (0x01) even when the U64E firmware returned a short or empty OPEN_TCP response — i.e. no real socket was actually opened. The C64 then commits to a phantom socket (uci_socket_id = 0x00) and pushes data into a non-existent connection. Combined with the unbounded spin in uci_wait_idle (see companion issue), this manifests as a 600+ s test hang.
Discovery context
Surfaced during validation of c64-test-harness issue #90 (JC-000/c64-test-harness#90). The downstream tools/uci/test_https_local.py test wedges for 600 s while the C64 pushes a 137-byte ClientHello to socket_id=0. The Python listener never receives a SYN, but the C64 thinks it's connected and is dutifully writing bytes via SOCKET_WRITE.
Full forensic analysis: JC-000/c64-test-harness#90 (comment)
Specific code
src/net/uci/net.s:447-464 — net_tcp_connect@tc_no_err. After uci_read_resp_bytes populates uci_socket_id, control unconditionally falls through to set net_tcp_state = UCI_TCP_CONNECTED, regardless of whether the read succeeded with a non-zero socket_id.
- Healthy run:
uci_socket_id = 0x06→ CONNECTED is correct. - Failing run:
uci_socket_id = 0x00→ CONNECTED is wrong; the C64 proceeds as if it has a socket.
Evidence (from c64-test-harness #90 artifacts)
Three independent failing-run dumps in /tmp/uci_https_debug/* all show identical post-wedge state:
{
"tls_state": "01 (CLIENT_HELLO assembled)",
"net_tcp_state": "01 (CONNECTED)",
"uci_socket_id": "00",
"tcp_recv_head/tail":"0/0",
"server_result": {"listening": true, "client_addr": null,
"error": "TimeoutError: timed out"}
}The listener saw client_addr=None after 600 s. No SYN ever arrived. Yet the C64 wrote 227k UCI register accesses and the full ClientHello to a phantom socket.
The PASS run (sibling test test_https_print_body on same device, same session) shows uci_socket_id = 0x06 (real socket) and the listener received the SYN at ('10.43.23.81', 52436). The two outcomes differ only in whether the firmware honored the OPEN_TCP — i.e. they are non-deterministic outcomes of the same code path.
Suggested fix
After uci_read_resp_bytes, validate that uci_socket_id != 0 (or whatever sentinel matches the firmware's "no socket" return) before setting CONNECTED:
@tc_no_err:
lda uci_socket_id
beq @tc_no_socket ; firmware returned no socket
lda #UCI_TCP_CONNECTED
sta net_tcp_state
rts
@tc_no_socket:
lda #UCI_TCP_FAILED ; or appropriate failure constant
sta net_tcp_state
sec
rtsCallers (tls_handshake, etc.) can then check the state and propagate the error cleanly instead of wedging on a phantom SOCKET_WRITE.
Severity
High — converts a transient (and apparently non-deterministic) firmware OPEN_TCP failure on U64E 3.14d into a 600 s test hang, blocking CI on the UCI suite. Combined with the unbounded uci_wait_idle spin (companion issue), the failure mode is silent rather than loud.
Activity
fix(uci): bounded uci_wait_idle + connect short-read detection + Labe…
feat(crypto): c64-x25519 v0.4.0 optional sibling integration (Phase C…