in reply to
Re: Typeless Relational Database
in thread Typeless Relational Database
As for type conversions ... if you're doing type conversions, either your schema is wrong or you're not using it correctly. You should never ever have to use a type conversion, not even once. And, yes, that's a hard rule.
That's going a bit too far, don't you think? I'll try to think of a really good example... What about something like this, maybe:
select ... from table where table.day_in_month = to_number(to_char(sys
Should I not be allowed to represent a day-in-month as a number? Should I, instead, be forced to use some sort of crazy "magic-number" reference month and year to store my day-in-month column, so that I can store it in a date datatype, and be able to do purely (but still crazy as all get-out) date-arithmetic to compare dates to the day-in-month? Imagine that the day-in-month is a piece of data used in representing a "recurrence" data-structure (like, "run this report on the 5th day of every month"), so I'm actually NOT talking about something that is a date... I'm talking about something more abstract, which can most easily be represented by type-casting (on top of some other logic).
Here's another kind of crazy (but real! I swear!) example: Another, really powerful use of type-casting in SQL comes up in a very fast (but admittedly ugly) way to pull data in one field by correspondance to data in another field. By that I mean, say, give me the X that corresponds to the min(Y). The specific case when I've used this with is trying to pull the earliest created value of something (say FOO is a number, for this example, and you want the FOO corresponding to the min(CREATED)):
min(to_char(CREATED, 'yyyymmddhh24miss') || FOO)),
If that's a little hard to understand, that's not too surprising, as it's something analogous to the GRT, but for seeking a min or a max, rather than for sorting (as the GRT is just a special case of the ST for sorting). That is: pack the carrier data (CREATED) together with the payload data (FOO), in such a way as comparisons of the packed carrier+payload correspond to any comparisons made against the naked carrier (i.e. if rowX.CREATED < rowY.CREATED, then to_char(...rowX...) < to_char(...rowY...), accordingly). Then, use comparisons made against the packed carrier+payload combo, which will yield back the data in which you're interested, but as packed carrier+payload. And then, finally unpack the payload from the carrier (i.e. the to_number(substr(...)) construct).
Anyway, I know it's esoteric, but it is real, and it's WAY more efficient than the more purist SQL:
where CREATED = (select min(FOO) from table where ...)
particularly if there's some interesting stuff in that "..." part of the query (which has to be repeated in the sub-query). (Oh, and yes, I did purposefully gloss over a few details in that example, that weren't vital to getting the point accross.)
Not an editor command: Wq