Broken version comparison on Anaconda.org

[ileasile]

It's the reposting of this issue

I've created a question on SO several months ago, but haven't received any answers.

Current Behavior

Different "latest versions" are reported for this package:
https://anaconda.org/jetbrains/kotlin-jupyter-kernel:

Shield from shields.io says that version is 0.9.0.6:

Shields.io makes a request to https://api.anaconda.org/package/jetbrains/kotlin-jupyter-kernel and gets

"latest_version":"0.9.0.6"

in the response.
But in the same response:

"versions": [
    "0.6.3.10", 
    "0.7.1.10", 
    "0.7.2.10", 
    "0.7.3.10", 
    "0.7_SNAPSHOT", 
    "0.7.29_add_teamcity_integration", 
    "11", 
    "0.7.12_add_teamcity_integration", 
    "0.7.17_add_teamcity_integration", 
    "0.7.dev18", 
    "0.7.dev19", 
    "0.7.dev21", 
    "0.7.dev22", 
    "0.7.12", 
    "0.8.0", 
    "0.8.dev23", 
    "0.7.15.64.dev1", 
    "0.7.40.145", 
    "0.8.0.2", 
    "0.8.0.7", 
    "0.8.0.10", 
    "0.8.0.25", 
    "0.8.0.28", 
    "0.8.1.74", 
    "0.8.1.75", 
    "0.8.1.76", 
    "0.8.1.78", 
    "0.8.1.83", 
    "0.8.1.98", 
    "0.8.1.114", 
    "0.8.1.118", 
    "0.8.1.120", 
    "0.8.1.121", 
    "0.8.2.1", 
    "0.8.2.2", 
    "0.8.2.4", 
    "0.8.2.5", 
    "0.8.2.53", 
    "0.8.2.61", 
    "0.8.3.1", 
    "0.8.3.64", 
    "0.8.3.122", 
    "0.9.0.3", 
    "0.9.0.5", 
    "0.9.0.6", 
    "0.9.0.7", 
    "0.9.0.14"
  ]

So... It's very strange

Steps to Reproduce

Open the link above

Expected Behavior

All the latest versions shown are the same and should be 0.9.0.14 at the moment of this issue creation

[jameslamb]

(came here from conda-forge/lightgbm-feedstock#59 (comment))

I'm not sure where the code that populates that part of the UI (or the latest version returned by e.g. https://api.anaconda.org/package/jetbrains/kotlin-jupyter-kernel) lives, but will just note.... 0.9.0.6 is lexicographically greater than 0.9.0.14.

"0.9.0.6" > "0.9.0.14"
# True

So I suspect the code populating latest_value was maybe just sorting the versions as strings, instead of as package versions.

[mattkram]

Since it appears the initial issue has been resolved based on conda/infrastructure#943 I will close this.