I helped finding and fixing [Issue 3917] unit tests failing - default model not initialized. It is very good that ArgoUML has unit tests. Without them I think that release problems would increase a lot. Near release almost everyone who cares will be a bit stressed and the possibility for errors will increase. For this we have the automated unit tests, acting as a safety net.
I should state that ArgoUML developers don't do as much as they should test driven development and test driven bug fixing. This is a practice that isn't enforced by the leader and by the commit rules. But, given time it might be...
Now, imagine that it was. Imagine that commiting a change in the production code should be done with a companion test or at least should be covered by tests. How difficult it would be to verify and automate this? We would need to mark commits to repository as atomic, make a complete build with unit tests executed with coverage and check that the commit is according to the coverage criteria. The commit is rolled back if it is sub-standard, being the commiter warned about this by email.
This requires atomic commits, something CVS does not have, but, Subversion does. It requires integration between the tools used in this scenario: version control system and build tool; build tool and coverage tool; build tool and coverage tool results details related to a set of specific changes; build tool and email system; mapping between version control system user IDs and users details (i.e. email).
Looking into the above requirements, I think that almost everything is possible with open source tools. It would require heavy build knowledge and a dedicated server with rollback rights, but, it is certainly possible.
One problem is the dependencies between one commit and the following. One much more important problem is the total lack of human review in all this!
No, in the end I think this is a bad idea. We should educate persons to follow good practices and then measure. Not put a machine enforcing this!