I have a small patch for cryptlib 3.2 to avoid a problem with handling
EOF on Unix. Here's the explanation I sent to Peter about the problem:

  readSocketFunction() (in misc/net_tcp.c in 3.1, and io/tcp.c in 3.2)
  has code that looks like this (paraphrased):

      /* We've got data waiting, read it */
      bytesRead = recv( stream->netSocket, bufPtr, bytesToRead, 0 );
      if( isSocketError( bytesRead ) )
          ...
      if( bytesRead == 0 )
      {
          if( isRestartableError() )
          {
              assert( !"Restartable read, recv() indicated no error" );
              continue;
          }
          ...

  Where isRestartableError() checks "errno == EINTR || errno == EAGAIN".
  The problem is, when recv() returns 0 on EOF, it didn't modify errno,
  and there is in fact no error; but the loop continues to select and
  recv nevertheless, and the function never returns.

The patch just removes the spurious check, which works fine on Unix.

--- io/tcp.c~	2005-04-26 10:54:51.153077738 +0530
+++ io/tcp.c	2005-04-26 10:18:18.939992468 +0530
@@ -1669,12 +1668,12 @@ static int readSocketFunction( STREAM *s
 			   uses a valid bytes-received value to indicate an out-of-band 
 			   condition that should be reported via an error code ("There's 
 			   nowt wrong wi' owt what mithen clutterbucks don't barley 
-			   grummit") */
+			   grummit")
 			if( isRestartableError() )
 				{
 				assert( !"Restartable read, recv() indicated no error" );
 				continue;
-				}
+				} */
 
 			/* Once this Winsock bug hits, we've fallen and can't get up any 
 			   more.  WSAGetLastError() reports no error, select() reports