[969] in Kerberos_V5_Development
On testing configurations...
daemon@ATHENA.MIT.EDU (epeisach@MIT.EDU)
Mon Jan 22 10:13:42 1996
From: epeisach@MIT.EDU
Date: Mon, 22 Jan 1996 10:13:37 -0500
To: krbdev@MIT.EDU
With all the possibilities of configuring krb5, I was wondering how
extensive the testing should be, at least from simply compiling
everything....
For instance, I can see the following series of options, a few of which
are incompatible with each other, but the general idea:
cc vs gcc 2
ccopts=-g, , -O 3
with & without shared libraries: 2
kdb_db: unspecified, db 2
aname_db: unspecified, db 2
with krb4, without krb4, with krb4 & kdb4 3
----------------------
product= 144
The reason I have the various ccopts is to weed out compiler
optimization bugs, etc....
Still 144 is rather large, which subset would really be a good set to
test?
Some of the testing is just to make sure that configure DTRT with the
multitude of options and need not be tested under every OS once you have
verified that the options work and are compatible with each other.
If you are concerned with simply compiler bugs, then you can reduce
testing by knowing that you are essentiall repeating a test. I.e. if you
use the berk_db code, once you test it with gcc and say -O for the kdb
database, you likely don't need to test it explicitly for the aname
database.
So any idea what a proper and manageable subset would be?
Ezra
