System V/AT v2.4 Impressions and Bug Report

[Michael J. Young]

I haven't seen much activity here of late.  Has anyone installed the
2.4 upgrade to System V/AT?  Here are my impressions of it along with
a few bug reports:

The new kernel has support for ESDI and RLL drives, although you have
to use the linkkit to get it.  Since I'm still stuck with ST506 controllers,
I haven't been able to try it out.  But I'm sure glad to see it there!

There is a new virtual console "utility" called "vcon" which eliminates
the need to create a getty for each virtual console.  Vcon is just run
once from inittab and attached to /dev/console, and it spawns gettys when
you hit the appropriate hot keys.  It's pretty nice, although there are
some strange behaviors that I don't like, and it's not configurable.
More on that later.

The hotkeys can be reconfigured using the new "mapkey" utility.  In addition
to defining a string that is sent each time a key is pressed, you can now
cause a signal to be sent to a process, or switch to any arbitrary console,
etc.  Not a bad idea, but it has problems too (below).

The disk driver seems to be better. The two-drive problem hasn't shown up
since I installed 2.4, and I'm beating 2 drives pretty bad.

The floppy driver has been improved.  Now you can access two floppies
simultaneously.  Under 2.3 I used to see erratic behavior sometimes (like
failed accesses for no reason), but everything seems clean under 2.4.

The maximum user process size is now user-configurable.  Unfortunately,
it didn't seem to help running pathalias, I still get a core dump when
I try to run it on the entire map database.  I've set the process size
to 4MB and the swap partition to 9000 blocks (I know that's less than the
recommended, but it should be enough for a single large process shouldn't
it?), but pathalias still fails when it tries to grow beyond 1MB.  Has
anyone else gotten pathalias to process the entire world?  How much
process space/swap space are needed?

There are supposed to be some fixes to the 287 support, but I haven't tried
it out yet.  There don't seem to be any substantial improvements to the
compiler, even though they issued a new SDS.  I sure wish they'd find
someone who can fix that stupid compiler.

New Bugs:
1.  Contrary to the release notes, PgUp and PgDn still seem to be
    defined backward.  It's easily fixed by putting a couple of "setkey"
    calls in an rc.d script.

2.  With the new kernel, when getty on a modem line exits (like when
    going from multi-user to single-user mode or complete shutdown), 
    DTR goes away, but then comes back!  It remains on forever!  It
    doesn't seem to cause any problems, but it was a little disconcerting.

3.  I still think the ansi terminfo entries are goofed up.  They still
    mess up jove, and even less had problems with it.  I went back to
    my (munged) 2.3 entry and everything is dandy.  I suppose the real
    problem could be in jove, etc., but I doubt it.

4.  Sometimes the keyboard locks up (scroll lock light on) and there
    doesn't seem to be a way to free it up short of reset.  Even Ctrl-Alt-Del
    stops working.  It's happened to me three times in four days, usually when
    I've accidentally hit more than one key at a time while using CTRL-S
    and CTRL-Q.  This never happened under 2.3, even though I wasn't
    any better a typist back then!

5.  Sometimes for no known reason "ixon" processing gets turned off and
    CTRL-S/CTRL-Q no longer work.  Typing "stty ixon" restores it, but
    it happens often enough to be bothersome.  I assume it has something
    to do with the previous bug as well.

6.  Control key processing has changed.  CTRL-SHIFT-2 (CTRL-'@') no longer
    emits a NUL (0x0), but a '@'.  In fact, under 2.3 CTRL-2 emitted a
    NUL.  Now it emits a '2'.  There no longer seems to be a way of getting
    the keyboard to emit a NUL.

7.  The "mapkey" utility has some bugs.  First of all, entering an
    illegal (unsupported) key name causes a core dump.  Secondly, using
    the form:
    	mapkey {keyname}
    causes an ioctl error, even though it is listed in the online man page
    as a legal construct (it should display the definition for the key).

8.  The "vcon" utility likes to assume that when you switch to a new
    virtual console you want to log in as the same user that you used
    on the previous console.  For example, say you're logged in as "guest"
    on virtual console 1 and you switch to console 2 for the first time.
    Vcon will automatically log you in on console 2 as "guest".  I'd
    rather just get a new login prompt.  The only way to get a login
    prompt is to first switch to an existing console that has a login
    prompt.  Cute,  but no thanks.

9.  A more serious problem with vcon is that creating new consoles is
    dangerous.  Creating too many (~4) new consoles too quickly causes
    a double fault panic.  You can do this by hitting Alt-F2, Alt-F3, etc.,
    in rapid succession.  This is a documented "feature".  Waiting
    until each new console is done being created before creating a new
    one seems to avoid the problem.

All in all the upgrade was an improvement, but not as much of one as I
would have liked.  I'm not as interested in new features like "vcon"
as I am in having a working compiler.  The new disk drivers are a welcome
addition, though.

Has anyone else installed the upgrade yet?  What do you think?
-- 
Mike Young
Software Development Technologies, Inc., Sudbury MA       Tel: +1 508 443 5779
Internet: mjy at sdti.sdti.com                 UUCP: {harvard,mit-eddie}!sdti!mjy