Re: Bug#698164: mandate unique package names in Debian Java policy
On 16/01/13 09:05, Eric Lavarde - Debian wrote:
> Hi,
>
> Daniel Pocock said:
>>> B. I doubt that such a badly named package would be of enough interest /
>>> quality for Debian packaging (but I might be wrong, I don't know any
>>> example)
>>
>> There are examples like this. It has been argued by some developers
>> that to compile using some toolchains (e.g. Blackberry builds done on
>> Windows), they have to shorten the package names, otherwise they end up
>> with filenames that are too long
> And who are we, packagers, to decide that this argument is wrong? And
Some of us might be wearing the upstream hat as well
And as packagers, we can always fork a project if the situation is
appropriate, and then rename the Java package prefix appropriately
> again, perfect example: if we change the package name on Debian, programs
> built on Debian wouldn't work on Windows and vice versa, and possibly
> neither on Blackberry which would make the principle of a toolchain quite
> absurd, wouldn't it. Wonderful result for a language supposedly
> portable...
Not quite... if upstream goes with the rename, other users are likely to
update their references to those classes too. I'm about to do this in
the gmetric4j repo, for example.
Here are some examples of packages that I have first hand experience
with as upstream, forking or using:
https://github.com/ganglia/gmetric4j/tree/master/src/main/java/ganglia
(import ganglia.*)
https://github.com/privatewave/zrtp-java/blob/master/src/zorg/ZRTP.java
(import zorg.ZRTP)
>>
>> Having this in Debian policy would help underline best practice and show
>> people evidence why they shouldn't deviate from it.
> I quickly checked other Debian language policies (Perl, Python) and
> couldn't anything even similar to this proposal, which confirms me in my
> opinion that policies are (rightly IMHO) about packaging not about
> programming best practices.
Yes, but programs (and libraries) do have to live in the packaging system.
Compared to C or Perl, Java has a vast number of library jars that get
used together in different combinations.
Even if this doesn't become mandatory policy, could we add a note about
it and a link to the JLS comments into the FAQ for Java packages,
encouraging people to check namespaces are somewhat unique and sensible,
even if conformance to best practice is not mandatory?
>>
>>> Conclusion: I vote against + close the bug.
>>
>> Can you leave the bug open so other people can comment for now? It is
>> not RC after all
> Sure, I meant to say that I vote to close the bug. Unless someone
> disagrees, I would suggest to leave it open until next week (gives one
> more week-end for people to react) and then close it (if votes continue to
> go in the same direction, of course).
Reply to: