On 9/27/07, Rupert Smith <[EMAIL PROTECTED]> wrote: > > Ok, the 'integrationtests' module, is for Java client against *any* broker > or *any* client in *any* language. The name may be a bit misleading, but > that is what that module is for. That is why the interop tests are there, > and if they are run in-vm, then they don't interop anything. You could run > them in-vm, but it would be pointless.
In that case integration tests is some what misleading. If named interop tests then this would have been more clear. 'systests' is for Java client + Java broker only. The java messaging system > in isolation from any other implementation. So we can move the current tests under client and broker over here as this what it does currently. I hardly see a difference between the tests in client/broker and the systests. Unit tests under 'client' are for unit testing the client code only. In > fact, they do not fit that intention and instead most of them are more > like > the tests under 'systests'. More correctly, they could be moved there. So > far, no-one can be bothered to do that, as it hasn't been a problem. It is a problem as people continue to write and run the wrong kind of tests under client and broker. You need proper unit tests to cover small areas of code/functionality ex a method or a few methods in isolation. This is the purpose of unit testing. What we have now is functional testing and is best achieved if moved to systests. We need proper code coverage to ensure we don't break functionality when we introduce new code. Run a code coverage tool and you will see what I mean. Rupert > > On 27/09/2007, Arnaud Simon <[EMAIL PROTECTED]> wrote: > > > > On Thu, 2007-09-27 at 10:20 +0100, Rupert Smith wrote: > > > Integration tests and client unit tests are quite dissimilar. > > > > > > The integration tests are aimed at interop. The client unit tests are > > > supposed to be pure unit tests for the client, but in fact they are > > > more like Java client + broker sys tests. > > > > This is the issue, the client unit tests are not pure unit tests for the > > client but more client+broker tests. Some of them are even more broker > > oriented. Those tests should therefore be run as part of testing the > > integration between a java client and a broker. That is to say that when > > those tests are run against a in-VM broker then they can be viewed as > > "unit" tests and when a remote broker is used as "integration" tests. > > Providing a configurable way of choosing the type of broker used to run > > those tests would provide us with both options. > > > > > Integration tests should not be run against an in-vm broker. > > > > Why not? > > > > > > > > > > > > > > >
