Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!husc6!rutgers!ames!ucbcad!ucbvax!decvax!decwrl!pyramid!uccba!hal!ncoast!allbery
From: allbery@ncoast.UUCP (Brandon Allbery)
Newsgroups: comp.emacs
Subject: Re: JOVE and/or GNU-MACS for 386/IX
Message-ID: <4096@ncoast.UUCP>
Date: Fri, 7-Aug-87 23:12:25 EDT
Article-I.D.: ncoast.4096
Posted: Fri Aug  7 23:12:25 1987
Date-Received: Sun, 9-Aug-87 11:19:10 EDT
References: <184@titn.TITN>
Reply-To: allbery@ncoast.UUCP (Brandon Allbery)
Followup-To: comp.emacs
Organization: Cleveland Public Access UN*X, Cleveland, Oh
Lines: 85

As quoted from <184@titn.TITN> by jordan@titn.TITN (Jordan Bortz):
+---------------
| Has anyone done a port of jove or gnu-macs for the 386 running Ineractive
| Systems 386/IX (V.3)??  I'm really looking for gnumacs, and would like to
| bench it (and use LISP funcs!), but am using jove now, which I compiled 
| on the 386.
| 
| It seems to work okay, except that running processes in windows doesn't
| work, and a make-to-buffer gives a Read Error 4.  Does anyone have
| the patches for jove?? 
+---------------


This bug in JOVE happens on any non-BSD system.  BSD has the property of
automatically restarting interrupted read()s on slow devices; others don't,
and the read() function returns -1 with errno == EINTR (4).

I don't have the fixes online.  Basically, however, I changed the call to
read() in _filbuf (? -- the function called by getc(); it's in fp.c) to
xread(), and added the following function (approximate; I don't know if this
one will work, but you get the idea):

#ifndef BSD

xread(fd, buf, nbytes)
int fd, nbytes;
char *buf; {
	long here;
	int cnt, rcnt, acnt;
	extern int errno;
	
	if ((here = lseek(fd, 0L, 1)) == -1)
		rcnt = 1;
	else
		rcnt = nbytes;
	for (cnt = 0; cnt < nbytes; cnt += acnt) {
		if ((acnt = read(fd, buf + cnt, rcnt)) == -1)
			if (errno != EINTR)
				return -1;
			else {
				acnt = 0;
				if (here != -1L)
					(void) lseek(fd, here, 0);
			}
		else if (acnt < rcnt)
			return cnt + acnt;
	}
	return cnt;
}

#endif BSD

The idea was to (1) force restarts on interrupted reads, and (2) do so in
an efficient way where possible (i.e. seekable files).  No checks for non-
pipes are done, since I have contempt for anyone who would C-X C-I /dev/tty.
Those who disagree can stick an isatty(fd) in the seekable check.

I suspect that the same has to be done for writes; I've had truncated writes,
and truncated file reads from before I did this.

One "bug" I've noticed, which I don't think is from my code:  we have an
early LSRHS version of Jove on ncoast, and it displays each line from a
C-X C-E as make outputs it.  The new version of Jove on tdi2 waits for
1024 bytes before displaying it; I consider this unfriendly, as I prefer
the proof that it's actually doing something.  (There may be another bug
in the fact that Jove works fine on tdi2 (a System V site) with my change
as posted above, but on ncoast the new Jove always gives me:

JOVE CRASH!!! (code 1)
Your buffers have been saved...

I traced this to random read(0, ...) calls in random places returning EOF
when it wasn't expected.  I don't know if this is a bug in Jove (unlikely,
or it'd happen on tdi2 as well; they use the same configuration, modulo
user-written memcpy()) or in System III (or Plexus Sys3; likely, given
the nonsense in both Plexus Sys3 and Plexus Sys5...).

I'll see about posting my actual changes.  BTW, the total set of changes
I made includes code to attach to /etc/avenrun's shm segment (see article
X-Archive: comp.sources.misc/8708/1) to support the mode line %l parameter.
-- 
 Brandon S. Allbery, moderator of comp.sources.misc and comp.binaries.ibm.pc
  {{harvard,mit-eddie}!necntc,well!hoptoad,sun!cwruecmp!hal}!ncoast!allbery
ARPA: necntc!ncoast!allbery@harvard.harvard.edu  Fido: 157/502  MCI: BALLBERY
   <<ncoast Public Access UNIX: +1 216 781 6201 24hrs. 300/1200/2400 baud>>