J1939SoftSocket: expand edge-case unit tests, soft↔native interop coverage, and pgn filter

[Copilot]

• Investigate CI failures
• Fix flake8 E501
• Fix TX race condition in J1939SoftSocket close() drain loop
• Fix _tx_start_bam/_tx_start_rts race: set tx_state before CAN send so close() drain loop doesn't break early when can_socket has per-frame delay (SlowTestSocket)
• Add more unit tests using TestSocket, SlowTestSocket, and USBTestSocket
  • pgn filter for RTS/CTS session (matching PGN delivered)
  • pgn filter for RTS/CTS session (non-matching PGN silently dropped)
  • pgn filter: 17-bit PGN (data_page=1) matched correctly
  • send() on closed socket returns 0 without raising
  • ABORT from sender resets RTS/CTS RX session; fresh session accepted
  • Duplicate TP.DT sequence number triggers ABORT
  • SlowTestSocket: BAM RX through serial buffer (internal _mux() path)
  • SlowTestSocket: BAM TX with 1ms per-frame delay
  • SlowTestSocket: BAM RX with 30 background frames filling serial buffer
  • SlowTestSocket: RTS/CTS receive with CTS response over slow CAN socket
  • SlowTestSocket: soft-to-soft BAM through slow socket
  • SlowTestSocket: listen_only — RTS/CTS passively reassembled without CTS/ACK emitted
  • SlowTestSocket: pgn filter — non-matching BAM dropped, matching BAM delivered through serial buffer
  • SlowTestSocket: RTS/CTS TX — sock as sender; peer responds CTS+ACK; validates full TX state machine
  • USBTestSocket: BAM receive through USB hardware FIFO
  • USBTestSocket: multiple sequential sessions survive background-frame FIFO drain between sessions
• Verify 185 non-root J1939 tests pass with 0 failures
• Verify no new flake8 issues introduced

[polybassa]

@copilot Analyze the J1939 SoftSocket implementation from BenGardiner:j1939_scanning https://github.com/secdev/scapy/pull/4961/changes#diff-5d6504c729d8bfec31ff30aec28e76f32ae70eba914b2338540fe2b13a221f34

Check for useful modifications

[Copilot]

@copilot Analyze the J1939 SoftSocket implementation from BenGardiner:j1939_scanning https://github.com/secdev/scapy/pull/4961/changes#diff-5d6504c729d8bfec31ff30aec28e76f32ae70eba914b2338540fe2b13a22...

Analyzed PR secdev#4961. Summary of what BenGardiner's implementation adds over ours, and what we incorporated:

Already in our implementation (confirmed equivalent):

• listen_only parameter – identical concept, fully implemented and tested
• Timeout extension factor (TP_DT_TIMEOUT_EXTENSION_FACTOR = 10) – we have _J1939_TP_DT_TIMEOUT_EXTENSION = 10 with the same total-elapsed-time logic; I clarified the comments to accurately describe the wall-clock ceiling approach (commit 63faffb)

Incorporated from BenGardiner's approach:

• pgn filter parameter (rx_pgn in his code) – added pgn_filter to J1939TPImplementation and pgn to J1939SoftSocket; pgn=0 accepts all PGNs (identical semantics to his rx_pgn=0); filters single-frame, BAM, and RTS/CTS sessions. Three new tests added (commit ea28d5a).

Not incorporated (out of scope / too complex for this PR):

• Configurable CMDT block size (bs parameter) – requires multi-block CTS state tracking
• J1939-81 address claiming (name parameter) – separate feature, significant complexity

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• https://api.github.com/repos/secdev/scapy/pulls/4961/files
  • Triggering command: /home/REDACTED/work/_temp/ghcca-node/node/bin/node /home/REDACTED/work/_temp/ghcca-node/node/bin/node --enable-source-maps /home/REDACTED/work/_temp/copilot-developer-action-main/dist/index.js (http block)
  • Triggering command: `/usr/bin/python3 python3 -c
    import urllib.request

Try different URL formats for BenGardiner's PR

urls = [
'REDACTED',
'REDACTED',
]
for url i iptables -w -t security ndor/bin/bash -nxv` (http block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[polybassa]

@copilot a few CI jobs fail because of the J1939 SoftSocket. fix them

[polybassa]

@copilot Continue

[Copilot]

@copilot Continue

Fixed in commit ae5bf26.

Two root causes identified and fixed:

1. TX race condition (caused 5 of the 8 CI failures — all "J1939SoftSocket TX → NativeJ1939Socket RX" tests): send() enqueues the message but close() was immediately setting closed=True and cancelling the TX scheduler before _tx_poll had a chance to dequeue and transmit it. Fixed by:

   • close() now waits (up to 2 s) for tx_state to return to IDLE and the tx_queue to drain before cancelling scheduler handles.
   • send() now cancels the pending TX poll and reschedules it with timeout=0, so the message is dispatched almost immediately rather than waiting up to 5 ms.

2. Flake8 E501 (code health check failure): the type comment on J1939SoftSocket.select() was 121 chars. Added # noqa: E501 to suppress it.

[polybassa]

@copilot add more unit tests, using testsocket and slowtestsocket

[polybassa]

@copilot continue

[Copilot]

@copilot continue

Added 5 more tests in commit c7cb328, bringing total non-root J1939 tests from 180 → 185 (all passing):

• listen_only via SlowTestSocket – RTS/CTS session passively reassembled; verified no CTS/ACK emitted
• pgn filter via SlowTestSocket – non-matching BAM silently dropped through serial buffer; matching BAM delivered
• RTS/CTS TX via SlowTestSocket – sock as sender over slow socket; peer thread responds CTS+ACK; validates full TX state machine
• USBTestSocket: BAM receive – frames buffer in the USB hardware FIFO until select() drains it
• USBTestSocket: sequential sessions – background frames between two sessions drained correctly by the CAN receive loop