I've heard the argument below often, but I've used the import pkg.*
convention on many projects before without any such troubles. In
particular, the Farrago project has a jillion dependencies on
third-party libraries. And when I upgrade one of them, I commonly have
to deal with semantic mismatches, but only very rarely with namespace
collisions, which are very easy to detect and resolve anyway. I don't
see why it would be any different for your intrepid JGraphT user
charging off into JDK 1.6 or a new release of JGraph: the real pain
during an upgrade always comes from semantic changes.
Interestingly, the Sun JDK source code uses a mix of the two conventions.
Let's leave this up to the maintainer's preference; when the time comes
for me to hand off the torch, the next maintainer can review this
convention. Or if users start swearing at us, then we'll review it
sooner. :)
JVS
Barak Naveh wrote:
> I still think the explicit list of import is preferable. To be
> argumentative, modern IDE's do indeed show where the class is coming from,
> but they also do clever folding :)... One can automatically unclutter the
> import list by letting the IDE fold it (I think it's even the default in
> Eclipse)...
>
> But now seriously; I have a better reason.
> Import such as java.util.* "contaminates" the namespace with unrelated
> classes that might cause name collision. To demonstrate how this
> contamination can cause problems, let's suppose that Sun adds a
> java.util.UndirectedGraph interface in jdk 1.6, and let's also suppose, that
> JGraphT, for good reasons, does not immediately upgrade to 1.6 (we know it
> took a while to update to 1.5).
>
> Now, we, in JGraphT, program against 1.5 so everything compiles cheerfully,
> passes the unit tests, and we're happy and proud. An enthusiastic developer
> who has already rushed to use 1.6 downloads JGraphT, attempts to compile it
> and... starts swearing... The import section of, say, AbstractBaseGraph,
> contains:
>
> import java.util.*;
> import org._3pq.jgrapht.*;
>
> which leads to a name collision -- the code doesn't compile.
>
> If a name collision with Sun sounds imaginary, a collision with JGraph
> wouldn't be that hard to imagine. Anyway, the use of wildcards for imports
> (regular or static) can cause problems that happen only in the "field" and
> can easily escape detection and correction in the "lab". After painfully
> experiencing such problems, one is compelled to decide to never again use
> wildcards for imports, static or otherwise. And if the clutter could be
> easily handled by the IDE, the decision comes with no sacrifice...
>
> Would you reconsider?
>
> Barak
>
>
>
>
|
