[Seaside] GOODS design question

[Brian Murphy-Dye]

Avi, this helps a lot and sounds pretty exciting! GOODS handling all of 
the referential integrity junk is definitely nice. Given your example, 
I would first need to

      (db root at: 'studentByName') put: Dictionary new

right? (Or perhaps 'BTree new'.)

I currently have a tangled web of 20,000 objects which I had tried to 
store in a GOODS database a few nights ago and it crashed the db. Do I 
need to add and commit a few objects at a time and slowly build up to 
the full 20,000 objects? Or perhaps GOODS doesn't handle 
SortedCollections?

Brian.


On Jun 16, 2004, at 3:28 PM, Avi Bryant wrote:

>
> On Jun 16, 2004, at 2:05 PM, Brian Murphy-Dye wrote:
>
>> I'd like to try using GOODS, but don't know how to design an object 
>> database. For example, in a relational database used for school 
>> course scheduling, there is basically a one-to-one mapping between 
>> the classes implemented in Smalltalk and the tables in the 
>> relationship database (with the addition of tables for the 
>> many-to-many relationships). But if you do something like 'db root 
>> at: 'Micky Mouse' put: aStudent' in GOODS, all of Micky's classes 
>> will be stored as well (assuming there is a 'classes' instance 
>> variable in the Student class), and then all of the students in those 
>> classes. Is this acceptable or should I design my classes so they 
>> don't contain other objects?
>
> That's totally acceptable.  Indeed, that's the whole point: that you 
> can use normal references between objects, and the usual collection 
> classes, and so on, without having to worry about joins.  One tradeoff 
> is that you have to managed indexing yourself - a more realistic 
> version of your example might be
>
> (db root at: 'studentsByName') at: 'Mickey Mouse' put: aStudent
>
> where you might also maintain studentsByStudentNumber, coursesByID, 
> and so on.  You'll generally need to use one of those indices to find 
> the main object you're interested in, and then follow references to 
> other objects from there.
>
> I should also point out the BTree package on SqueakMap - if you were 
> going to have, for example, 5000 students indexed by name, you would 
> be much better to use a BTree for the index than a normal, hashed 
> Dictionary.  It would be nice to see some other data structures (Trie 
> comes to mind) if anyone has time... the performance characteristics 
> of an OODB are (perhaps obviously) much closer to disk than to RAM, so 
> all those data structures you learned about in school but rarely use 
> are suddenly important again.  ;)
>
> Avi
>
> _______________________________________________
> Seaside mailing list
> Seaside at lists.squeakfoundation.org
> http://lists.squeakfoundation.org/listinfo/seaside
>