Hi Stephane!

ducasse <ducasse at iam.unibe.ch> wrote:
> Hi goran (avi)
> 
> I and Avi have discussed and made a first sketch on a "deployment plan"
> > for the infrastructure needed. This includes putting an improved PI 
> > into
> > the update stream, proposing to move Monticello into "Base" (!) and to
> > deploy SM2.
> 
> I'm really interested into that.
> By the way does SM2 have dependencies between SM2 packages.

No. Not yet. But as I have explained before :-) there is a dependency
scheme planned and coming.
SM2.0 will bring us Accounts, Package releases, co-maintainers, a
simpler synch mechanism (a slight hack for now), a local package cache
(and in 2.1 a server cache), a personal file area on the SM master and a
whole bunch smaller stuff I don't remember right now.

The important thing in the above is of course package *releases* which
means the map keeps track of all releases made of a package. Thus you
can install whichever version you like, and this means you can build
"static loadscripts" where "static" means they will produce the same
result even if they are installed in 5 months and there have been many
more releases of the packages involved. In SM1.x you can only install
"what happens to be the latest release of package X".

> Monticello should have the same dependencies between packages

Well, Monticello and SM handles two different kinds of packages. Sure,
sometimes an SM package *is* a Monticello package, but it is not
generally the same thing. So I am not sure that we are talking about the
same mechanism.

IMHO I would like to handle dependencies *between packages* on the SM
level. Avi will probably start making something on his own for
Monticello, and he is of course free to do that - he might as I said
even have other problems than SM tries to solve - and I only "pray to
God" that it will not conflict with what I have planned in SM2.x. :-)

But I think Avi has a pretty good picture of what is planned in future
SM.

> > Avi is listening for PI/Monticello requests so if there are any that 
> > you
> > think would be very good in the focus of TFNR, don't hesitate to talk 
> > to
> > Avi.
> 
> I discussed a lot with avi off list on related topics. I think that we 
> agree that
> we need to fix also the fact that certain run-entities do not have 
> enough information.
> So the natural follow up would be to go in the direction of
> 	- having a run-time package knowing its class and methods

Eh, PIs do that, right?

> 	- having a run-time class knowing its package

Well, again - PIs do that. If we have PIs for all classes - and that is
what TFNR task 3 (Ned, Brian, Adam) is producing.

> 	- having a run-time class knowing from which pakage it is  extended
> 	- a compiled method knowing its package (if its package is the same 
> than the one of its => no extension else extension)
> 	- having global variable declaration
> 	- having class declaration (different than subclass:.....)

Well, these last I can't comment on really - I assume that some of the
enhancements being done in PI have to do with this.
 
> This way we could still have the complete tools working on run-time 
> entities and not declarations
> but still support the tools such monticello. This requires to fix some 
> tools to produce the right information.
> 
> Stef

regards, Göran