deploying jars with classifiers appears to be broken

[cprice404]

Many of our projects include 2 jars; a 'main' jar and a second jar with a 'test' classifier that contains test utility code for use by downstream libraries.

Historically we've had no troubles with this approach and it's worked fine with clojars. Here's an example of an artifact that was successfully published with this pattern:

https://clojars.org/repo/puppetlabs/trapperkeeper-webserver-jetty9/1.5.2/

However, we tried to do another release of this library today and, although lein reported a successful deploy and logged information about sending up the test jar, the repo directory does not contain the test jar:

https://clojars.org/repo/puppetlabs/trapperkeeper-webserver-jetty9/1.5.3/

Subsequently, I tried to do a deploy of a 1.5.4-SNAPSHOT artifact for the same project. It also showed the test jar being sent up to clojars in the lein output, but it ended up in an even weirder state:

https://clojars.org/repo/puppetlabs/trapperkeeper-webserver-jetty9/1.5.4-SNAPSHOT/

in this deployment it appears that the test jar was uploaded but the filename it was given is the filename of the main jar; so, the main jar has the test content, and the "real" main jar just doesn't seem to be in the repo.

Has anything changed lately that would affect the deployment of jars with classifiers?

[danielcompton]

Yep, we just had a deploy which changed validation of uploads https://groups.google.com/forum/#!topic/clojars-maintainers/Y4ToszKujDU, this is almost certainly the issue. Paging @tobias.

[cprice404]

Any chance that could be rolled back? We're working on a release of some downstream projects that are blocked by this, so we're starting to consider some drastic options that will make us sad :)

[tobias]

@cprice404 - sorry for the trouble - I'll take a look right now to see if this would be a straightforward fix before considering rolling back.

[cprice404]

@tobias thanks! much appreciated.

[tobias]

@cprice404 I'm trying to recreate locally. What's the lein command you would use to deploy with the classifier jar included?

[tobias]

Nevermind - I have built a simpler app that demonstrates the issue. Working on a fix now.

[tobias]

@cprice404 - I've deployed a fix to production - can you deploy 1.5.4-SNAPSHOT again? If that works, a release deploy will work, and I can then delete 1.5.3 for you if you like and let you redeploy it as well.

[cprice404]

@tobias awesome, thanks so much, will test now.

[cprice404]

I redeployed 1.5.4-SNAPSHOT but it's hard to tell at a glance if that worked because there are a lot of files in there now.

So I did a deploy of a 1.5.4-testclojarsfix-SNAPSHOT:

https://clojars.org/repo/puppetlabs/trapperkeeper-webserver-jetty9/1.5.4-testclojarsfix-SNAPSHOT/

It still appears to just have one jar, the pom, and the checksum files, unfortunately.

[tobias]

Ah, darn. My fix won't work for SNAPSHOTS, since the version in the pom doesn't match the version of the file (since it's timestamp-based). Let me take another crack at it.

[tobias]

Alright, another fixed deployed, and I was able to deploy both snapshot and non-snapshot releases that have classifiers of a test app. @cprice404 - can you try another snapshot please to confirm?

[cprice404]

yep, testing now

[cprice404]

@tobias that looks like it worked for the SNAPSHOT. I'm going to just try doing a 1.5.4 release now.