Re: [gobo-eiffel-develop] cluster structure of bridge pattern

[Alexander K. <kw...@ah...>]

Andreas Leitner wrote:

> foo ... main cluster of GOBO-sub-library "foo"
> foo/bar ... "bar" is a (interface-)subcluster
> foo/foobar ..."foobar" is another (interface-)subcluster
> foo/impl ... "impl" contains all bridge-implementations
> foo/impl/win32 ... contains the win32 implementation
> foo/impl/win32/bar ... implements the "bar" interfaces for win32
> foo/impl/win32/foobar ... implements the "foobar" interfaces for win32
> foo/impl/unix/ ... contains the unix implementation
> foo/impl/unix/bar ... implements the "bar" interfaces for unix
> foo/impl/unix/foobar ... implements the "foobar" interfaces for unix

> For several reasons (some implementations might be mutaly exclusive or
> depend on external references you cannot satisfy) it must be easily
> possible to select an arbitray number of implementations. This means
> that if the above foo example is in a XACE file called foo.xace, you may
> not want mount the whole cluster, as this would automatically mount all
> implementations.
> 
> This is easily _not_ possible with the above cluster structure for
> "foo". It is possible with the original cluster structure of eXML
> thought. Here is the same library "foo" with the eXML (or Vision2 - to
> be exact) cluster structure:
> 
> foo ... main cluster of GOBO-sub-library "foo"
> foo/interfaces ... all interfaces subclusters go in here
> foo/interfaces/bar ... bar is an interface-subcluster
> foo/interface/foobar ... foobar is an interface-subcluster
> foo/impl  ... all implementations go in here
> foo/impl/win32 ... the win32 impl
> foo/impl/win32/foo .. the "foo" impl for win32
> foo/impl/win32/foobar .. the "foobar" impl for win32
> foo/impl/unix/foo ... the "foo" impl for unix
> foo/impl/unix/foobar ... the "foobar" impl for unix 
> 
> Here the user mounts (remember a mount, also mounts the sublcusters)
> 
> foo/interfaces
> foo/impl/${PLATFORM}

Like Eric I see an additional level "interfaces" as artificial.
In fact, if the cluster description file is designed correctly,
every implementation cluster will refer to the corresponding
interface cluster, so

 foo/impl/win32/foo mounts foo/foo
 foo/impl/win32/foobar mounts foo/foobar
 foo/impl/unix/foo mounts foo/foo
 foo/impl/unix/foobar mounts foo/foobar

So, it would be enough for the user to mount only one
cluster, namely

 foo/impl/${PLATFORM}

All the interfaces will be mounted automatically.

I would also move PLATFORM/COMPILER-dependent
directories down the hierarchy:

 foo/impl/foo/win32
 foo/impl/foo/unix
 foo/impl/foobar/win32
 foo/impl/foobar/unix

and hide the platform dependencies in the descriptions
of the clusters  foo/impl/foo and  foo/impl/foobar so that
the client should not worry about implementation issues
at all. The user just mounts

 foo/impl

or more selectively

 foo/impl/foo

Regards,
Alexander Kogtenkov
Object Tools, Moscow