search

smallrye / smallrye-common

Common utilities for SmallRye
Apache License 2.0
21 stars 24 forks source link

3.6.0.CR1 considered newer than 3.6.0 #267

Closed gsmet closed 10 months ago

gsmet commented 10 months ago

Hey,

I tried to use smallrye-common-version for our release scripts and things went wrong because 3.6.0.CR1 is considered newer than 3.6.0.

I switched back to using Maven Artifact as these two versions are sorted properly with ComparableVersion.

The algorithm in MavenVersionScheme is not exactly trivial so I will refrain from breaking things :).

/cc @dmlloyd

dmlloyd commented 10 months ago

For background on this problem, see:

Long story short, the Maven version scheme has changed incompatibly at least twice. I considered the possibility of having multiple schemes depending on the Maven version in question but I think what I will do is just reimplement it to support the current scheme and hope they don't change it again.

gsmet commented 10 months ago

Well, at least in this case, they have fixed something so I think it's a good thing.

dmlloyd commented 10 months ago

My solution is to stop using the maven-artifact scheme and use the maven-resolver one instead (they're different!), and add a few more tests to ensure we're fully consistent. Part of this involved a minor internal enhancement to make it easier to compare against "zero" versions (because in Maven, versions like "CR1" or "beta" are considered to be less than zero).

gsmet commented 10 months ago

Hmmm, I'm using maven-artifact ComparableVersion to work around the issue so I'm surprised you have to switch to something else?

dmlloyd commented 10 months ago

The problems you're going to run into with maven-artfact are bit more obscure, like differing precedence of - versus . separators, and the fact that maven-resolver also treats _ as a separator. The problems I was running into were a result of twisting the code around to account for all of the various internal consistencies in maven-artifact (to give an example, between versions 3.8.6 and 3.8.7, the sort order of e.g. 0.x vs 0-x swapped). By going with the simpler model, there are vastly fewer bugs that need to be worked around, which means a greater likelihood that the sorting rule ends up being useful.