[Geotools-devel] FM Hacking - Guidelines for GeoServer next

[Jody G. <jga...@re...>]

Cory Horner wrote:
> Hey Jody, Justin,
>
> I had a peek at FM this morning... took a few stabs at fixing it, but I 
> have a question which i'll phrase in the form of a class:
>
> org.opengis.feature.AttributeFactory?
>   
Now called FeatureFactory. Since this is a feature model, and since we 
now have more then attributes I decided not to play around anymore.

Code is committed, although I had to battle a few conflicts, my own 
fault for keeping the code so long - thanks for releasing a new geoapi 
jar BTW.

So far in talking to Justin I have found a few ground rules that when I 
screw up has me fail a code review. These are hard to capature .. but I 
gotta try anyways.

#1 Builder vs Factory:
- XFactory does what it does *exactly* with no logic
- XBuilder maintains state and can do the logic you wanted to put in the 
factory

#2 DataStore and its Factories:
- AbstractDataStore is focused on "simple" features
- AbstractDataStore "owns" types and thus *needs* a SimpleTypeFactory in 
order to create them
- FeatureSource "owns" data and thus needs a SimpleFeatureFactory in 
order to create data

#3 Iterators and FeatureReaders/FeatureWriters
- FeatureReader and FeatureWriter are no longer a focus of interaction
- Iterators "create" content, each pass through the data will need *its 
own* FeatureBuilder

All of Data basically fails Guideline #3 and will need to be fixed. You 
can see this failure when ever you see a FeatureBuilder (or any Builder) 
passed around as a parameter between classes, or stored for reuse). This 
is hard to test as the problem will only occur when the DataStore is 
being used by two threads (or if a FeatureCollection is being used by 
two threads). Let me capture that as a guidelines.

#4 Builders should only ever be a local variable.

Once again sorry for being slow on this kind of writing,
Jody