Hello Ludo, welcome back! On 03/06/2017 06:00 PM, Ludovic Courtès wrote: > Is it 100% reproducible if you run: > > ./check-guile 00-repl-server.test > > from Guile’s build tree? > > This test uses a Unix-domain socket, which on the Hurd means that > /servers/socket/3 (I think?) must have the right translator on it. > > 00-socket.test also uses Unix-domain sockets. Does it pass? > > Looking more closely, it might be that one of the hunks of the patch > below solves the problem. Could you try and report back? > > (Looking at > , I > think ECONNRESET is more appropriate than ENOTCONN in the second case.) > > HTH, > Ludo’. > Since the last email I sent, I found out that I was getting ENOTCONN only after the second time I was running the test, and every time after that, unless I delete /tmp/repl-server. The error you get the first time you run the test is FAIL: 00-repl-server.test: repl-server: simple expression - arguments: (expected-value "scheme@(repl-server)> $1 = 42\n" actual-value "scheme@(repl-server)> While reading expression:\nERROR: In procedure fport_fill_input: Resource temporarily unavailable\nscheme@(repl-server)> While reading expression:\nERROR: In procedure fport_fill_input: Resource temporarily unavailable\nscheme@(repl-server)> While reading expression:\nERROR: In procedure fport_fill_input: Resource temporarily unavailable\nscheme@(repl-server)> While reading expression:\nERROR: In procedure fport_fill_input: Resource temporarily unavailable\nscheme@(repl-server)> While reading expression:\nERROR: In procedure fport_fill_input: Resource temporarily unavailable\n$1 = 42\n") I am testing with "GUILE_LOAD_PATH=. ./guile-test tests/00-initial-env.test tests/00-repl-server.test" and it's 100% reproducible if you delete /tmp/repl-server after each run. 00-socket.test passes each time successfully. Your patch doesn't solve the first error. Trying to debug the problem using rpctrace causes both tests to end with unresolved test cases. I am attaching the rpc-trace output. Manolis