What is the usually procedure for such things? Wouldn't this cause eval errors if somone is still using one of these versions? Maybe we should add aliases which throws an error instead? Not sure.

Yes, this will definitely cause eval errors if someone is using those. However, nobody should be running such ancient software so I consider this a good thing.

Having aliases is probably quite dangerous because one needs to perform some manual steps when upgrading.

Result of nixpkgs-review pr 105781 run on x86_64-linux 1

Having aliases is probably quite dangerous because one needs to perform some manual steps when upgrading.

I would still prefer if we add aliases for those and warn about the upgrade steps.

I am strongly against aliasing the old versions. I do think aliasing would make it very easy to shoot oneself in the foot. Let's say someone updates to a newer NixOS version but doesn't read the release notes too carefully. If he deploys the new version, this will silently upgrade Kafka and very likely cause downtime to the Kafka cluster. As Kafka is often used as highly-available message broker and as essential part of distributed systems, this needs to be avoided at all costs.

I see two feasible options:

1. Remove the old versions without aliases.

2. Keep the old versions.

If we go with 2, we probably should mark all old versions as insecure as I highly doubt that Apache provides security updates for e.g. version 0.9 (released 2015).

I think there is also precedent for having to manually update the package name, specifically postgresql. They are also somewhat similar as both Postgresql and Kafka usually require manual steps when performing upgrades between major versions and that they both often are critical services.

I mean adding aliases that throw an error that we removed the package rather than redirecting to a later version. This is nicer as it tells the end user what changed rather than failing to evaluate their config.

Ah sorry, I completely misunderstood 😄

I added a simple error message about having to upgrade.

@phile314 I think the eval error is fixed in master. Could you rebase? Otherwise the PR looks good to me.

@ofborg eval