bug#49496: 28.0.50; process-tests/fd-setsize-no-crash/make-network-process hangs on Cygwin

bug#49496: 28.0.50; process-tests/fd-setsize-no-crash/make-network-process hangs on Cygwin

[Ken Brown]

The hang can actually be triggered by evaluating the following, extracted from 
the test in the subject:

(setq socket-name "/tmp/socktest/socket")
(delete-file socket-name)
(make-network-process :name "server"
                       :server 10
                       :buffer nil
                       :service socket-name
                       :family 'local
                       :coding 'no-conversion
                       :noquery t)
(make-network-process :name "client"
		      :service socket-name
		      :family 'local
		      :coding 'no-conversion
		      :noquery t)

The reason for the hang is that Cygwin does a blocking handshake in 
accept/connect to exchange credentials.  This can hang in certain situations in 
which connect is called before a listening socket is ready to accept connections.

It is possible to work around this problem by disabling the credential exchange 
as in the following patch:

diff --git a/src/process.c b/src/process.c
index b8c3e4ecfb..01548c8ab5 100644
--- a/src/process.c
+++ b/src/process.c
@@ -3355,6 +3355,12 @@ connect_network_socket (Lisp_Object proc, Lisp_Object 
addrinfos,
  	      xerrno = EMFILE;
  	      continue;
  	    }
+#ifdef CYGWIN
+	  /* Avoid possible hang in connect/accept.  */
+	  if (family == AF_LOCAL && p->socktype == SOCK_STREAM
+	      && setsockopt (s, SOL_SOCKET, SO_PEERCRED, 0, 0) < 0)
+	    report_file_error ("Cannot disable credential exchange", Qnil);
+#endif
  	}
         if (p->is_non_blocking_client && ! (SOCK_NONBLOCK && socket_to_use < 0))

But this seems too drastic just to fix a potential problem that doesn't seem to 
occur in practice, but only in the contrived situation of 
process-tests/fd-setsize-no-crash/make-network-process.  So my preference is to 
simply skip that test on Cygwin.

OK?

In GNU Emacs 28.0.50 (build 9, x86_64-pc-cygwin, GTK+ Version 3.22.28, cairo 
version 1.17.4)
  of 2021-07-09 built on moufang2
Repository revision: 68276f6d30bbdc09cc26fb49d7f0c3aa4bce35f2
Repository branch: master
Windowing system distributor 'The Cygwin/X Project', version 11.0.12011000
Configured using:
  'configure 'CFLAGS=-g -O0''

Configured features:
ACL CAIRO DBUS FREETYPE GIF GLIB GMP GNUTLS GSETTINGS HARFBUZZ JPEG JSON
LCMS2 LIBOTF LIBXML2 M17N_FLT MODULES NOTIFY GFILENOTIFY PDUMPER PNG
RSVG SOUND THREADS TIFF TOOLKIT_SCROLL_BARS X11 XDBE XIM XPM GTK3 ZLIB

bug#49496: 28.0.50; process-tests/fd-setsize-no-crash/make-network-process hangs on Cygwin

[Eli Zaretskii]

> From: Ken Brown <kbrown@cornell.edu>
> Date: Fri, 9 Jul 2021 14:01:33 -0400
> 
> It is possible to work around this problem by disabling the credential exchange 
> as in the following patch:
> 
> diff --git a/src/process.c b/src/process.c
> index b8c3e4ecfb..01548c8ab5 100644
> --- a/src/process.c
> +++ b/src/process.c
> @@ -3355,6 +3355,12 @@ connect_network_socket (Lisp_Object proc, Lisp_Object 
> addrinfos,
>   	      xerrno = EMFILE;
>   	      continue;
>   	    }
> +#ifdef CYGWIN
> +	  /* Avoid possible hang in connect/accept.  */
> +	  if (family == AF_LOCAL && p->socktype == SOCK_STREAM
> +	      && setsockopt (s, SOL_SOCKET, SO_PEERCRED, 0, 0) < 0)
> +	    report_file_error ("Cannot disable credential exchange", Qnil);
> +#endif
>   	}
>          if (p->is_non_blocking_client && ! (SOCK_NONBLOCK && socket_to_use < 0))
> 
> But this seems too drastic just to fix a potential problem that doesn't seem to 
> occur in practice, but only in the contrived situation of 
> process-tests/fd-setsize-no-crash/make-network-process.  So my preference is to 
> simply skip that test on Cygwin.
> 
> OK?

Yes, OK.  But please explain there in a comment why we skip that, with
a reference to this bug.

(FWIW, I consider the process-tests suite to be not useful enough, in
that it many times tests behavior that is only guaranteed to happen on
GNU/Linux, and is not really required, nor should be expected, in
portable Lisp programs.  So it doesn't surprise me that some of the
tests fail on some systems.)

bug#49496: 28.0.50; process-tests/fd-setsize-no-crash/make-network-process hangs on Cygwin

[Ken Brown]

On 7/9/2021 2:18 PM, Eli Zaretskii wrote:
> Yes, OK.  But please explain there in a comment why we skip that, with
> a reference to this bug.

Done.  Closing.