OK, Myrna, thanks, very helpful.
David
Myrna van Lunteren wrote:
On 6/14/05, *David Van Couvering* <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
No, I just ran it once with the default (which I guess is jdk14). I had
been under the mistaken impression that derbyall included some jdk13
tests.
David
Hi David,
The test harness assumes the lowest level of jvm anyone will run with is
jdk13.
However, there are tests that test functionality which wasn't there with
jdk13 spec, so those will be excluded from running with jdk13 (they'll
have a runwithjdk13=false in some property). In addition, there are some
tests that mostly run the same with jdk13 & later, but have some minor
differences which are still correct behavior. For those tests, we have
jdk14 (or other, as applicable) canons.
So, if you run with jdk14 only (which most of us do), you might miss the
fact that a test has a minor difference with jdk13.
The way to prevent messing up this mechanism, is to verify that any
masters you update do *not* have a jdk14 specific canon. If they do, you
ought to run with jdk13 & update the 'master', as well as run with jdk14
& update the jdk14 (functionTests/master/jdk14/) canon.
Or like mamta says, run with both jdks if you're modifying behavior that
would have different effects with jdk13 & later.
Also, if you add a new test or subtest that tests out functionality that
wasn't there with jdk13, then you should see if you can just make the
test skip with jdk13 (runwithjdk13=false & runwithibm13=false) or if you
can add a canon. Again, the master ought to be the jdk13 result, and any
canons should go in their specific spots.
Myrna
