[rfoo]

But then another customer still wants gamma@1, they are allowed to do that! But they also want your new features. So now you have to maintain two branches, which I hope we can agree: it is an extra burden.

This is unavoidable if we are talking about FOSS, people should be able to do whatever they want, and they do. A company has an advantage here: you can install company-wide rules and culture to make sure people don't do this. Which, in this case, happens to be: let's keep a single version of everything unless you have really good reasons.

[spion]

> But then another customer still wants gamma@1, they are allowed to do that! But they also want your new features.

In this case, you still have the option of working with them to help them migrate to gamma@2, if the cost of maintaining gamma@1 is indeed too high and would negatively impact you in serving other customers. This was the original premise, wasn't it - upgrading all your dependants when you upgrade your library? That's still an option. The point is - you have more choices. And you can also help customers one by one - you don't have to do it all at once

I will agree though, restricting choices helps when the company is finding difficulty in aligning incentives through communication. But you do give up a lot for it - including ability to move fast and avoid stagnation.

[rfoo]

Ah, ok. I see where you are coming from :)

From what I saw I'd say it's exactly opposite: allowing multiple versions actually means "make teams able to choose for stagnation". And because we are lazy, we certainly do! There is a non-trivial amount of people who believes "if it ain't broken don't fix it". I can work with them to migrate them over, but they might not want to do so! In this case, a hard "bump versions or die" rule is a must.

Maybe if you work in a small group of great engineers you don't need to set such rules and you can move even faster, but I unfortunately haven't found such a workplace :(

> you don't have to do it all at once

Yes. Nobody should do it all at once. Making "bump versions or die" compatible with incremental adoption is slightly harder (see sibling threads for how it's done). Still worth it I'd argue.