Re: What databases have taught me

Laurent Bossavit wrote:
> Marshall,
>
> > It depends. If we imagine the same query existing in two places
> > in the code, and imagine what could make us want to change it
> > in the future, do we imagine that we'd necessarily change *both*
> > together or might we just want to change one?
>
> I'm wondering how this argument goes if we replace "two" by "ten" or
> "one hundred".

Hardly any. Given the same query existing N places in the code. We can define an equivalence relation such that when we need to change the query at location i, we also need to change it (in the same way) at location j.

> What is it that caused us to need the same query to appear at two
> different places - and how likely is it that the same forces will
> require that we have the same query in three places, then maybe four,
> etc. If we do have four or more instances of the same query, do we more
> easily imagine wanting to change all of them?
>
> Your question "how little code can there be such that we still call it a
> coherent abstraction" is interesting when you ask it about the original
> example:
>
> select name from employees where today()-birthday > X and city Y
>
> Wouldn't you say that the calculation "today() - birthday" calls for
> abstraction ?

No I definitely would not. I probably wouldn't object if you made one, but if you do, you haven't really done much of anything. You've made an abstraction that saves one subtraction, and that abstracts a definition that every 6 year old knows.

> It really means "age", so the code should say "age". The
> means of abstraction doesn't have to be a method - a view would work in
> the above.
>
> I would want to ask, too, what the condition "older than X and living in
> Y" means. It seems likely that it stands for something more abstract:
> say, "the employee is eligible for a particular healtcare plan".
>
> If that were the case, the answer to your original question would be
> "Change Both", because when the rules for eligibility changed, the
> change would need to apply to all places in the code where we compute a
> list of employees based on eligibility for that plan.

If the analysis works out that way, then yes. But it's hard to say in advance.

Marshall Received on Wed Jul 05 2006 - 17:06:15 CEST