My experiences

[utzoo!decvax!ucbvax!ucsfcgl!katzung]

I've brought up a lot of V7 software on an 11/34, and many of the problems
I've been reading about sound familiar.  Here are some of my experiences:
(Please excuse any redundancy with old news items, I don't get news on
the 11/34)

1.  C bug warning
	In rewriting our v6 vt11 driver for v7, I changed some pointers from
'int' to 'int *'.  I later discovered that the driver was writing data into
strange addresses.  It turned out that the compiler was multiplying the shift
count in some memory management register calculations by 2, causing a shift
by 12 instead of a shift by 6.  Some tests showed that a '>> 1' shift on,
say, a pointer to a 17 byte structure effectively became '>> 17'.  Admittedly,
this is a strange case, but the problem might conceivably crop up in other
places.

2.  Oddities in 'ps'.
	When I first got the system going, ps only printed about half of
the 'CMD' entries, the others being printed as '()'.  The problem was that
I had linked /dev/swap to the raw device instead of the block device
(causing in-core processes to print correctly, and swapped processes to
appear as '()').  The 'u_comm' field shouldn't need to be used unless your
system has commands that destroy 'argv' information.

3.  Overlayed Pascal
	I have overlayed pi (everything else fits), although I'm sure it
is not divided into the most efficient overlays possible.  However, it
has worked fine on the few programs I have run.

4.  Spell, awk, and f77
	I was able to shrink spell down small enough to fit by halving the
hashing table.  It runs quite well, but I don't know how badly the hashing
system has been degraded.  So far, I have not gotten far with awk, and
I have not even looked at f77.

5.  Various versions of ex
	In an effort to keep size down, I decided to turn off some of the
define options for ex version 2, but discovered that no makefile existed
for an overlay version (which was quite strange, since the supplied ex was
an overlayed version 2).  I converted the version 3 makefile, which produced
an EXTREMELY slow but functional version 2.  Profiling indicated that over-
lays were being switched some 2000 times for every ^F and ^B.  Changing over-
lays around improved things greatly, but multiple simultaneous edits dragged
the system to a crawl.  We are currently using a VERY modified version of
ex1.1 with an unbelievable history.  It's only saving feature is that it is
very fast, since it fits without overlays.  (Are we the only ones with a
problem?  I haven't heard any complaints in the news.)

Brian Katzung	ucbvax!ucsfcgl!ucsfpgs!brian