From: Frederic G. <go...@pu...> - 2003-02-24 21:03:10
|
> Again, I'm aware that databases make some things easier. What I'm saying > one hand and the Python DB-API + whatever DBMS (+maybe a pyblio-specific > generic DB layer) on the other. I guess the big difference between the > functionality of the two is that ZODB *may* make efficient searches > harder. Even that really isn't obvious to me, though -- if our object > model doesn't fit SQL + relational DB model well (hierarchical folders?), > it's possible that query scalability will actually be worse with a > relational DB (as well as requiring more mapping code). This is an issue > that I'm not well-informed enough on yet to know the answer, though (if > there is 'an' answer independant of the particular problem, which is > doubtful). :( You are right, the model won't fit as-is in a SQL model. This is also why we are spending more time describing the requirements and thinking about the data model in an abstract way. Then we'll try to find out an API that can support it and see how it must be adapted to fit actual back-ends (abstractions always leak...) There will certainly be some work to achieve a good mapping, it would be a waste of time to start it too early. Some spike implementations are interesting though, only if to check what can be done with the tools at hand. I've never used ZODB, I need to have a look at it. I know that Peter is inclined to use a DB3 backend, whereas I would li= ke to push my knowledge of PostgreSQL a bit further. Small backends (like gadfly, or maybe no DB at all) could be implemented as well.=20 As you can see, nothing is decided yet, and how could it be ? We are really exploring a wide area, but python is so well suited for these kinds of explorations that I think we might have a chance to build something _really_ good. Fr=E9d=E9ric |