[1428] in Kerberos_V5_Development
Current krb5 on Alpha
daemon@ATHENA.MIT.EDU (epeisach@MIT.EDU)
Wed Jul 24 15:00:16 1996
From: epeisach@MIT.EDU
Date: Wed, 24 Jul 1996 14:59:44 -0400
To: krbcore@MIT.EDU
I have checked in the changes required to get krb5 to work on the Alpha
again. The following still needs to be dealt with;
a) make check from top fails in lib/kdb/t_kdb.c...
b) Shared library building is broken in kadm5 - Tom is working on this.
The real problem on the Alpha stemmed from the why xdr_long is
implemented. xdr_long will only send 32 bits OTW. Now, if the
sizeof(long) > 4, additional code is put into place to see that the
higher 32bits are not set - which is good. The problem is that an
uninitialized field of a structure which normally was just sent OTW on
other platforms - could cause a failure on the Alpha. (I don't know why
the implementation uses longs for obj_attributes as there will never be
more than 32bits - but that's another issue).
I am changed the kadmin client code to zero out stack based structures
which of type kadm_principal_ent_rec - but I may have missed one.
I also did not do a totally complete check throughout the sources to see
if any were missed. So there may be potential pitfalls out there...
To make matters more amusing - the dejagnu test suite no longer has the
kadmin tests in there (except for the basics to add and entry and
get info) so I can't even tell if the password changing code would
work.
So I am a little concern about the lack of proper testing of the new
kadmin system, but unless someone has the time to redo those tests - I
cannot guarantee there will not be other hidden demons waiting to be
hit.
Ezra
