Niclas Hedhman wrote:
On Friday 09 February 2007 08:05, Carlos Sanchez wrote:
ok, so there's no "transitive dependency" thing like Maven has,
Speaking of that; IMHO If artifact A dependsOn artifact B which dependsOn
artifact C, then the transitive dependency should not provide C's classes to
A at compile time. Reason; If A uses anything directly in C, it would mean
that if a later version of B stop using C, A becomes 'unusable'.
Fast forward to OSGi; If you had the somewhat broken way of Maven allowing B
to automagically supply classes to A, then if one replaces B with another
implementation then A breaks. Therefor, A imports the C packages explicitly,
if they are needed, but if it is only internal use of C in B then it would
not require A to know about C.
Correct. The transitive type of dependency in OSGi is captured by "uses"
constraints, but this is still done in terms of packages. For example,
bundle A can export package 'foo' and export (or import) package 'bar'
and can declare that 'foo' uses 'bar'. This tells the framework that if
bundle B imports 'foo' and is wired to bundle A's 'foo', then if bundle
B imports 'bar' it must also get that from bundle A (or from the same
place as bundle A if a imports 'bar').
Thus, bundle A can inform the framework about transitive dependencies so
that it can resolve package dependencies in a consistent way for a set
of bundles. But, as Niclas points out, even then this does not
automagically give bundle B access to these "used" packages, bundle B
must still explicitly import them to have access to them.
-> richard
Try to stop thinking "bundles" and start thinking "packages". Bundles are
typically not referenced at all, especially not the generic case for the
maven-jar-plugin, which is what I am hoping you are working on.
Cheers
Niclas
