PEER_PORT shouldn't be ShortTag about opentracing-java

[wu-sheng]

Thanks for your report. I just pr the codes to fix this.

[yurishkuro]

  public static void main(String[] args) {
    long port = 33000;
    short shortPort = (short) port;
    System.out.println(shortPort);
  }

// Output: -32536

[wu-sheng]

@yurishkuro The output is -32536, it do not equal the expected value.

[wu-sheng]

-1 - this is API breaking change. A cast to short still preserves the full 16 bits of the port.

I understand your concern. But it really should not add this cast operation, when you set a port....

[yurishkuro]

Well, semantically it's the correct representation, e.g. InetSocketAddress defines the port as int, but in the comments limits the range "value is between 0 and 65535" (i.e. unsigned short). I agree that int would've been more convenient in practice, but it will be an API breaking change which will break existing instrumentations.

[wu-sheng]

Agee. Maybe I can change the pr, add some comments of PEER_PORT to describe this. In cases others have the same questions. Thoughts? @yurishkuro

[yurishkuro]

Yes, definitely makes sense to clarify the comments for that tag.

If others are ok to break backwards compatibility, I can be out-voted :-) After all, Java is still <1.0

[wu-sheng]

I hesitate about breaking backwards compatibility too. I will cc the pr to other contributors, see what happen next? OK?

[wu-sheng]

Just done. #114 (comment)

[wu-sheng]

Merged already.