I agree that #2 is a good option. Version parsing is used in various places, so I think we should still use a number. It also makes it clearer to consumers of the spi to know where to pull in dependencies.
On 10/17/18, 4:10 PM, "Jonathan Hurley" <[EMAIL PROTECTED]> wrote:
It looks like Ambari’s trunk pom.xml has not been updated with a proper version in at least 4 years. It currently still lists trunk as 184.108.40.206-SNAPSHOT: https://github.com/apache/ambari/blob/trunk/pom.xml#L24
This poses several problems as we try to make our artifacts more 3rd-party friendly since it becomes ambiguous what other maven projects are actually depending on. I think that we need to change our process to keep this version updated with every release of Ambari. Other Apache projects seem to do this (I took a look at Nifi, Hive, Storm, etc) and in each case, their trunk pom.xml was at least a minor version ahead of their most recent release branch.
I see three avenues for us here:
1. We can make trunk the next “known” version of Ambari. This gives us relatively fine grain control over snapshot artifact versioning, but also opens us up to problems when surprise versions show up. For example, up until a few weeks ago, trunk would have been 220.127.116.11-SNAPSHOT, however now that we have a 2.8 release off of trunk, we’d need to change this to 18.104.22.168-SNAPSHOT.
2. We can make trunk the next major version, so that it would currently be 22.214.171.124-SNAPSHOT. After 3.0 is released, then it would move to 126.96.36.199-SNAPSHOT even though we know that there are interim minor releases coming out (like 3.1, 3.2, etc)
3. We can abandon numbering for trunk and use something like TRUNK-SNAPSHOT. This also poses some ambiguity problems since you actually don’t know from which point in time the snapshot is really from. There’s also a problem with our regex parsing of the version if we switch away from the 4-digit scheme we have now.
I’d like to get some opinions on this topic. I personally think that #2 makes the most sense.