[FLINK-3560] [examples] Remove unchecked output of usage statement in examples

[tillrohrmann]

This PR effectively removes all unconditioned outputs of the usage message of the examples.

[vasia]

If we remove the default usage printing, I think it would be nice to make sure that all examples implement ProgramDescription. This way, we can get the usage information if we want to. Otherwise, we would get one message per default parameter. e.g. in the Kmeans example, we would get:

Executing K-Means example with default point data set.
Use --points to specify file input.
Executing K-Means example with default centroid data set.
Use --centroids to specify file input.
Printing result to stdout. Use --output to specify output path.

instead of
Usage: KMeans --points <path> --centroids <path> --output <path> --iterations <n>

[tillrohrmann]

That is a good point @vasia. Will add the missing ProgramDescriptions.

[tillrohrmann]

Added the missing ProgramDescriptions.

[StephanEwen]

I remember a discussion from a while back that we actually wanted to drop ProgramDescription.

It comes from the days when programs had to implement specific interface and it does not really go together very well with the idea of a simple static main-method as the entry point.

[uce]

I think Stephan has a point and we should maybe not encourage usage of ProgramDescription in the examples, because we advertise the simple main-method entry point in all the docs.

Other than that, I think it's a good idea to remove the unconditional usage printing.

[tillrohrmann]

ProgramDescription has not been marked as deprecated so far. Furthermore it can be used to show information about the program using bin/flink info and is not mandatory.

Admittedly, this feature isn't used by many programs but I think it does not hurt to keep and implement this interface, unless we plan to remove it soon. There was a discussion which also involved ProgramDescription. However, it does not look to me as if it converged to some real action points.

[fhueske]

I don't think the examples should implement the ProgramDescription interface. This might give the impression that implementing this interface is mandatory while it is a very optional thing.

I think it is not a problem that optional program parameters such as max-iterations won't be printed anymore. I would expect users to look at the code anyways to understand what is happening and there they will see the optional parameters.

[tillrohrmann]

Hmm but if it is nowhere implemented then we could directly remove ProgramDescription. I mean why keeping code in the base which is not used?

Furthermore, I don't think that it would harm if people implemented this interface. And if they look in the code, then they will also see that it's highly optional.

[StephanEwen]

From my side, +1 for removing the ProgramDescription interface.

I think it is a mismatch with the main() method based approach, because it needs to instantiate the class, while the general entry point is static. That does not go well together in my opinion. People will write programs with no default constructor. Previously, with the Program interface, such a constructor was also mandatory, it did fit better there.

[tillrohrmann]

That is a good point with the instantiation of the instances. I will update the PR accordingly.

[tillrohrmann]

Removed the ProgramDescription interface and rebase the PR on the latest master.

[uce]

Thanks, Till! I think this is ready to finally be merged. I will go ahead and do it.