Eunice bugs (confirm.&anal.)

[utzoo!duke!chico!harpo!mhtsa!ucbvax!sdcsvax!jmcg]

Subtopics:	future EUNICEs --- batch bug --- page-read-err bug

I called Dave Kashtan to ask about the bugs described below.  He was a little
surprised that I was blaming EUNICE for the problem--pointing out that
EUNICE runs entirely at user level.  Clearly HE hasn't seen these problems.

In reply to my suggestion that he include in the distribution some of the
sources for the UNIX programs that were changed in porting to EUNICE, he
described what he hopes to be the next distribution; it will essentially
be a complete distribution tape a la Berkeley's 4BSD with sources, makefiles
and all that jazz.  He didn't say anything about sources for EUNICE itself.
If he gets some help porting stuff over, it will happen sooner.
[porting frequently doesn't involve much more than checking that all
the file names generated have sensible translations as VMS file names.]

------------------------------------------

 Batch bug

We experienced a widespread failure of processes following submission of a
batch job that contained EUNICE commands.

Specifics: (I was trying to recompile uucp)
	Batch job defined 
		cc := $bin:cc cc
	submitted from system manager account (no BYPASS privilege).
	EUNICE resides on system disk.

Symptoms: Most users could not log in.  They would get as far as the
	"Welcome to VMS..." messages and nothing else would happen.
	Typing <cr> got back the Username: prompt.  Users who could
	log in found that their processes died suddenly when certain
	operations were attempted.

	System reboot was required.

Analysis: Something got the ACP for the affected disk very confused.  It
	was possible to use the 'dir' and 'dump' commands on the affected
	disk, but processes attempting to OPEN (ACCESS, etc.) files
	there were apparently deleted without any messages being put out
	from the exception/condition/exit handlers.  Other disks were
	unaffected.  Process creation and image activation (including
	Eunice images) appeared to be normal.  Submitting process was not
	affected until it attempted file access.  The batch job responsible
	for the mess left an incomplete .LOG file that stopped sort of
	the invocation of the Eunice command.  A different batch job with
	files already opened appeared to be unaffected.

	Alan Lesgold (lesgold at cmua) reported a very similar-sounding problem
	in a message to netnews dated 31 May.

-----------------

 page-read-err bug

Process failure following invocation of indirect command file containing
Eunice command.

Specifics: Same as above, except that I tried running the same .COM file
	from DCL as an indirect command file ('@') rather than SUBMITting it
	as a batch job.

Symptoms: Same as above, but first I got an error message complaining about
	Page Read Error followed by rather voluminous dumps of registers
	etc.  I'll try it again sometime when I won't disrupt the users
	and catch the results on a hardcopy terminal.

	Bill Mitchell (duke!unc!whm) reported something like this in a
	NET.eunice submission 12 Jun (unc.901?).

--------------------------------------------------------
				--Jim McGinness
				ucbvax!sdcsvax!jmcg