Unpublishing leads to test system crashs (4.3.1/4.3.2 )

[leoselig]

This refers to the issues deriving from #96
I'd like this to be an objective discussion (The unpublish note in the 4.3.2 release is not what I regard as objective)

If we have to expect that handling of the peerDependency again or any other SemVer violations this package simply has to be removed from our projects as it does not suite the requirements of a stable continous integration. That would be sad since it's a great plugin and we really enjoyed using it.

[leedm777]

If we have to expect that handling of the peerDependency again or any other SemVer violations this package simply has to be removed from our projects as it does not suite the requirements of a stable continous integration.

@leoselig If you need that level of stability for your test dependencies, I highly recommend you use npm shrinkwrap --dev to control them. SemVer isn't perfect. Any package can accidentally introduce compatibility breaking changes. It even happened with chai.

In general, there's not a great way to handle accidental breaking changes. Often times, reverting the breaking change is itself a breaking change. So do you violate SemVer again? Or do you unpublish the offending version? Or do you revert the change as a major rev bump? (Chai did it the last way, BTW).

There's no perfect answer. You are between a rock and a hard place, and you're gonna break some people. As a maintainer, you use your best judgement and hope you don't get too much grief for it.

[leoselig]

Thanks for the answer
I totally agree with you, trusting SenVer to keep your system stable is risky - I think I actually have to reconsider the strategy there
However, npm shrinkwrap wouldn't have worked in this case as far as I understand since the most important issue here that two versions where unpublished
I think against the latter there is actually no realistic protection which us why unpublish is strongly prohibited and requires a force flag

[keithamus]

Just happened upon this. The breaking change in Chai 1.10.0 was entirely my fault, and I apologise to anyone who had trouble by this. I also handled the transition badly, it took too long to fix between 1.10.0 and 2.0.0, so once again, I apologise.