Problems with XOpenDisplay

[mcr@ccs.carleton.ca (Michael Richardson)]

>   I think we never figured out the real source of the problem, but it
>   seems that XOpenDisplay will fail if any interrupt occurs while it's

  Probably the read()'s from the socket are not restarted if stopped
due to EINTR. This is not that surprising, since posting an alarm(2)
might be your only way to get out of a connection to an X server that
was taking `too long'. siglongjmp()'ing out of a system call is not
possible on some systems (I'm told). Still... I'm am a *little* surprised.

>   waiting for a response from the server.  We "solved" the problem by
>   blocking signals around the call.  [Self gets 50 or so timer

  Can you post a patch? I don't see any such code in the 2.0.1
source tree. (And to think, I just made clean this morning, thinking that I
wasn't going to have to rebuild again soon...)
  Across the campus backbone, to my poor, slow, NCD X terminal is
likely part of the definition of `too long' --- it also explains why
it takes two retries when going across scs.carleton.ca's net to a Mach
box, and why there is no problem when run on the console.