Re: SIGSEGV a bug or a feature?

Andreas Plesner Jacobsen <apj_at_daarligstil.dk> wrote in message news:<slrnc21g90.qs5.apj_at_slartibartfast.nerd.dk>...
> On 2004-02-03, Joel Garry <joel-garry_at_home.com> wrote:
>
> >> My only problem is: Oracle won't acknowledge that this is a bug. The
> >> claim is: This is an application sending us bad data, so we don't see a
> >> problem in handling this with a server-side crash (which gives the
> >> client a nice EOF on communications channel, but no good error message).
> >>
> >> So, what's your take on this? Isn't this a bug in Oracle? Should ANY
> >> invalid data be able to crash the Oracle process it's connected to?
> >
> > Send the data to metalink and crash _their_ server!
>
> I don't have sqlnet access to their database server ;)

As they say "first, you get the mule's attention." ;)

>
> > Which process are you referring to? If it is just a spawned beq they
> > might have a point, user written application bugs are out of their
> > purview.
>
> Yes, but really - should I be able to send data to a spawned session and
> force it to die with a segmentation violation. Shouldn't I get a more
> meaningful explanation than EOF on communication channel?
> I can't help but wonder how many more input validation errors there are
> in Oracle.

Yeah, you are right. There shouldn't be anything a user could do to sigsegv an _oracle_ process.

Can you give them a simple and easily replicable example using only Oracle and platform native software? It's been too long since I've DBI'd, but I seem to recall there are translators to C. You might ask the perlmongers, too.

jg

--
@home.com is bogus.
http://slashdot.org/article.pl?sid=04/01/29/1931227&mode=thread&tid=188&tid=97
http://imdb.com/name/nm0080049/