> BioMart has been designed in such a way that, the buffer for such > changes is the dataset configuration file, > which describes what _currently_ available in the database and > guarantees that anything in this file can be used to construct a valid > query. > These file were designed precisely for this purpose that the query > logic can be decoupled from evolving data.
Assuming that you always start from scratch that's great, the problem is that by its nature we store query specifications in the workflow. These specifications include filter and attribute names (how else could they do it?) and users expect that a query written against version x of database y shouldn't suddenly stop working. Our interface will pick up the new configuration sure, it's not hardcoded in our code, but that doesn't help a user who writes a workflow then comes back the next day to find that it randomly fails. You need to take the potential for stored queries into account. Tom
