Pushed a fix for this in v0.57.10 (not released yet, but the change is in). Want to loop you in on the diagnosis in case something else surfaces on your end.

What it was

A regression I introduced back in v0.54.7 when I fixed a similar-looking bug on Mystic/EleBBS.

On Mystic and EleBBS, the BBS hands the door a live TCP socket handle through DOOR32.SYS; the child process inherits the actual SOCKET. On exit, .NET's Socket finalizers were calling closesocket() on that inherited handle, which closed the BBS's own socket, sent a TCP shutdown to the remote user, and made the door unrelaunchable without a full reconnect. The v0.54.7 fix was to call ExitProcess() directly on Windows BBS exit, skipping .NET finalizers entirely so the OS just closes handles cleanly.

Problem: I guarded that on IsOSPlatform(Windows) only. Fired for every Windows BBS exit regardless of whether we were in socket mode or stdio mode.

For NFU + Renegade, we're in stdio mode; the child door doesn't own any socket. stdin/stdout are pipes owned by NFU, and NFU sits between us and the DOS BBS's FOSSIL driver. ExitProcess terminates us before .NET flushes stdout, so NFU sees a truncated stream instead of a clean EOF. The FOSSIL emulation gets stuck waiting on us, NTVDM locks up, Renegade hangs waiting on NFU, node dies until you reboot. Matches what you described exactly.

What v0.57.10 changes

New DoorMode.IsNativeSocketMode property; true only when we're in door mode AND stdio wasn't forced AND we actually have a valid inherited socket handle from the drop file. The exit gate is now:

if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows) && DoorMode.IsNativeSocketMode)
{
    NativeExitProcess(0);  // Mystic / EleBBS path - skip finalizers
}
else
{
    try { Console.Out.Flush(); } catch { }
    try { Console.Error.Flush(); } catch { }
    Environment.Exit(0);   // NFU, Synchronet stdio, MUD relay - clean EOF to parent
}

Mystic and EleBBS still get the native exit (they genuinely own a socket). NFU, stdio-mode Synchronet, MUD relay, Electron, and local testing all get a proper flush and a normal exit so parents get a clean EOF.

Need to ask you a favor

I can't reproduce NFU + NTVDM locally, so once v0.57.10 is out I'd really appreciate if you could test it on your Renegade setup. Specifically:

Launch the door via NFU, play for a bit, exit normally.
Confirm the BBS is still alive after the door exits.
Relaunch the door without rebooting the host, confirm it loads again cleanly.

If any of those three still fail, drop a log here or email and I'll dig in further. The fix is narrow and doesn't touch anything on paths that were already working, so regression risk on Mystic/EleBBS/Synchronet is low — but NFU is genuinely a code path I can only verify through you.

Thanks for sticking with this and for the detailed repro steps.

Uh oh!

Does not work with NFU and Telnet Door #75

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions