10 Jun
2011
10 Jun
'11
5:43 p.m.
> When working on eclipse, I discovered the
necessity to represents
> things like plugins or components which are not directly handled in
> Famix.
> It is important if we want to be able to deal
with real world cases.
I am not sure I understand. Could you be more
specific?
> May be we should consider introducing them as
Moose Entities?
FAMIX is meant to be extended. For example, to model
JEE systems, we
added entities like EntityBean or SessionBeans.
If I works with Eclipse project, at some point I will want to handle plugins as entities.
For example, if the application is very big, or if I want to think about the architecture,
I will want to work with plugins (components) rather than packages.
So now I want plugins in Moose.
Of course, I can extend the metamodel, but this means I have to think where to put it, may
be implement metrics, create the MooseChef queries, ...
Whereas by using MooseGroup I can instantly prototype components without changing Famix,
so I have some guarantee that things will just work.
Except that to work, many things require bidirectional navigation (from group to members
and members back to its group(s) ) ...
> Or may be not: I used Moose-Groups (of class
because one package can
> be spread over different plugins) and it is not that bad.
> the main issue, in my case, was that classes don't know to what
> group they have been assigned...
Groups are not meant to work bidirectionally. But,
perhaps it would be
an interesting addition to try to add this information.
> So computing dependencies could be very lengthy.
> I solved that by creating a dictionary on the side, but maybe we
> should think on a better solution.
> And then of course integrate this with all the
tools...
Interesting. I would be interested in hearing the
actual cases.
Two concrete cases that are cited above:
- when computing metrics on eclipse, I wanted the metrics at the level of plugins. Instead
of defining a FamixEclipsePlugin entity, I created each plugin has a class group, and
implemented the few metrics I needed on that (which required bidirectional navigation).
- we are supposed to look soon at a big eclipse "application" where the guy is
interested in looking at the dependences between plugins because there are too many
classes and packages to work at that level.
This implies having bidirectional navigation.
If DSM, metrics, MooseChef where already working on MooseGroups, I would have an instant
poor man component ...
- It could be used to define application layers and reason about them. We had the issue
recently for a paper, where we wanted to look at dependencies between a package and
itself, the package with its application, and the package with the underlying layer(s)
(e.g. a framework on which the application is defined). For this we had to hack some
dependency queries ...
- I could also maybe test various possible system decomposition using groups (somehow
similar to what Orion can do).
nicolas