On Dec 14, 2007, at 11:01 AM, Benson Margulies wrote:
We have about 10 copies of the hello world schema floating around the
tree.
The idea is that wsdl-first and code-first test schemas and
implementation should live in testutils. The problem, from my
jaundiced
point of view, is that we have been dumping things into there without
enough care in picking names.
I agree.
For example, someone put a copy of the hello_world 'greeter' schema
into
testutils, but modified it to use the XML binding.
In my opinion, it should have been renamed and renamespaced to in the
process. The current state is unproductive in two ways.
First, new people looking for a test case schema are prone to do
what I
did, which is see it, try to use it, and then belatedly run into
the XML
binding specification. And then have to back up and turn around.
Second, it turns the testutils into a confusing arena in which no one
really knows what we have for inventory.
I understand why the testutils stuff is there, but I don't like it.
The idea is to shorten build times, by putting a common set of
functionality in a JAR, and re-use to your heart's content.
My problem with that is re-use doesn't always work, so you get what
you observe -- lots of test-specific functionality in testutils.
(Look at hello_world_secure.wsdl, for an example).
What would be better (IMO) would be to possible put base schema and
logical WSDL types in testutils, and then have the actual tests
import these schema and WSDL from system tests, but only if it makes
sense. You'd still "pay" for code generation during the build, but
that's life in the fast lane. The current approach leads to spaghetti.
It looks to me as if another process is to take some example we get
in a
test case, and check it in, 'as is'. I can see the reason to do this:
ultimately, the user's test is the regression test. However, there are
all kinds of opportunities for unexpected classpath and namespace
conflicts this way. I wonder if we should 'pom' these regressions as
individual sub-projects of systests so that they ran in comparative
isolation from each other.
While I'm inventing work with no one to do it, I'd propose that:
a) all the services used in the demos should be in testutils,
verbatim,
and we should routinely run unit tests against those services.
Wouldn't there me a maintenance issue there? I'd rather see the
sample programs run as part of a build.
b) Some cleanup of the existing inventory would be helpful. I'll do at
least a little bit.
That would be good. I'd offer to help with the security stuff, but
I'm flat out right now.
c) Some wiki-writing explaining the inventory would be helpful.
RoTR would be good in this area.
--benson
