goran.krampe at bluefish.se wrote:
> We really should try getting that boat in the water though and I think
> we simply need to figure out how we "put PI instances into the image
> using the update stream". If we just solve that (should be easy, someone
> just have to do it) then we can start pumping out some PIs and we can
> easily integrate that with SM and BFAV.
> 
> If noone else does this I will. Now, how did this darn PI thingy work
> now again... ;)
I suggest we use PI subclasses instead of PI instances. I find them much
better for the following reasons:
- They are more straightforward to put into the update stream. ;-)
- They are more explicit than PI instances and thus easier to
understand.
- For some packages we need to have subclasses anyway because there is a
need for overwritten messages. Using PI subclasses for all packages
reduces the number of necessary concepts by half (from two to one ;-)
and is thus easier to understand. Think STTMPW.
- With PI subclasses there is a natural place to put things for a
package, should the need arise. Think package documentation. ;-)
- The PI subclasses would be a natural place to put code for package
initialization, reinitialization, loading, unloading.
- Monticello could be more easily used for versioning and merging
package meta-information.
- They would be very similar to ENVY applications and subapplications
and thus very easy to understand for former ENVY users.
- Porting code from and to VW/Envy and VA would be easier.

So, what do others think?

- Bernhard