Date:         Sun, 9 Nov 1997 20:26:05 -0500
Reply-To:     Jim Kennedy <72130.1210@COMPUSERVE.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Jim Kennedy <72130.1210@COMPUSERVE.COM>
Subject:      Re: Invalid file header
Content-Type: text/plain; charset=us-ascii

We've been experiencing the invalid header problem with
SAS 6.12, Windows 95, and Novell 4.11 for months.
SAS tech support is well aware of this problem and has been
incredibly useless.  The response to our last phone call
was "use your backup dataset" -- as if retrieving a backup
dataset and re-doing the latest data processing was now
standard practice.  We have lost many hours of work on
many occassions.
 
In our shop, the problem only occurs with certain workstations.
It never happens on our 3 Dells but happens on other computers
(notably Gateway).  We have expanded memory in the Gateways to
32 megs and rebuilt the hard disks from scratch and it still
happens.
 
I looked at the headers for good and bad files (using the old
DOS list.com program) and found that the files with header
errors had 133 bytes missing starting at byte 129.  It appears
that the rest of the file, including all the data, is ok.  I
wrote a program to cut the missing 133 bytes out of a good
file and paste them into a bad file.  When I did that, the
modified datasets need a password to open and, of course, I have
no clue what the password would be.  We have never used the
password option on any of our datasets.
 
So, my take on all this is that the header damage occurs in the
section of the header that handles the encrypted dataset password
and SAS has no interest in revealing any information about it or
releasing a fix.  My guess is that it is some type of weakness in
the SAS encryption algorithm -- they probably tried to be too
clever and lost reliability.  It is unfortunate that 12 megabytes
of data become unusable because of 133 bytes at the top.  I expect
that writing a program to fix the files would be easy if a person
knew what those 133 missing bytes did -- but, presumably, the
program might also allow people to bypass the dataset password
protection.
 
To minimize problems, we routinely keep at least two backup
copies for every important dataset and write a copy before
and after each editing session, sort, etc.  We still get the file
header errors, but can usually recover the data from the backup
versions and have not lost significant data in a couple of months.
It can take a lot of disk space to do this, and I'd hate to guess what
our cost in man hours has been for dealing with this.
 
 
Jim Kennedy
Colorado Prevention Center