From what I've traced this has something to do with the version algorithm implementation :
1. tokenize the string using "-" (dash) as the separator
2. checks if each token matches the regex pattern (see below)
3. the token ratio between "matching tokens" and "all tokens" must be 75% or more.

From the example it's only 50%.

We can loosen the rule a bit to return true if the first token is a matching token. thoughts on this ?

        "([0-9][_.0-9a-z]*)",
        "(snapshot)",
        "(g?[_.0-9ab]*(pre|rc|g|m)[_.0-9]*)",
        "(dev[_.0-9]*)",
        "(alpha[_.0-9]*)",
        "(beta[_.0-9]*)",
        "(rc[_.0-9]*)",
        "(debug[_.0-9]*)",
        "(unofficial[_.0-9]*)",
        "(current)",
        "(latest)",
        "(fcs)",
        "(release[_.0-9]*)",
        "(nightly)",
        "(final)",
        "(incubating)",
        "(incubator)",
        "([ab][_.0-9]+)"

attach patch for the proposed solution.

Is there logic somewhere in Maven to decide whether a version number is valid? Why would Archiva be more strict about this?

I've never had a build fail due to an invalid version number... it may not be able to find the dependency, and creative version number formats don't work well with ranges, but not invalid.

maven deploy plugin doesn't have this checking. it's only in archiva

added UT. THanks

Patch applied to trunk -r772151. Thanks Jan!

Leaving this issue for the documentation patch..

Updated documentation in trunk -r778123.