Re: ecc, error correcting code insertion

[bob@proulx.com (Bob Proulx)]

> (tried apt-cache search ecc, and a few other keywords)

You were probably thinking of the ecc package available from GNU
several years ago.  I had used it when one of our encrypting routers
developed a problem which would corrupt TCP data streams such as ftp.
You might think that TCP is always reliable but it is possible that it
won't be given bugs in implementations.  Of course we needed our data
to be correct after ftp transfer and ecc rescued us and worked well.

I was sorry to see the author pull the package from distribution.  We
used MD5 to do an overall end-to-end check to ensure data consistency
as I am the paranoid programmer.  One of the expected failure modes of
hamming codes is that too many bit errors can go undetected.  We were
routinely hitting the small number of errors in our transfers which
were correctable.  And in the case that ecc failed an end-to-end
signature check would detect that case.

The author apparently became very concerned about the possibility of
silent data corruption because of too many bit errors.  The program
did not include any end-to-end signature check to detect that case and
by itself was susceptible if used in isolation.  The package was
withdrawn from the GNU sites.  However it can still be found in many
archives by web search.  It was distributed under the GPL.

Bob


README file for the withdrawn ecc package:

                Withdraw of support for the ecc package
                =======================================

There is a fundamental bug in the ecc package (any version: 1.2.1 or 1.3.x).

In about 15% of the cases where there are 4 or more errors in the encoded
blocks, the ecc decoder would wrongly report that there were 3 or fewer
errors and that they had been corrected.  In actuality, not only were the
original errors not corrected, the decoder had inserted more errors.

Support for the ecc package is withdrawn until further notice.

If you are considering to incorporate the library in any applications,
please do not use it.  My sincere apologies for spoiling your plans.

Regards,
fclim@acm.org


-- 
To UNSUBSCRIBE, email to debian-user-request@lists.debian.org 
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org