[ regression ] Packages with same transitive dependencies no longer build

[stefan-hoeck]

Today I got a couple of mails from GitHub: Most of my packages' nightly builds failed. It looks like their (transitive) dependencies are no longer acceptable to Idris2. For instance: elab-util has a dependency on base >= 0.5.1, as does sop, which also depends on elab-util. Idris now complains that sop would depend on two versions of the same package.

Is this expected behavior? Pinging @Z-snails because this surely comes from #2469 .

Now, in case of sop, this could easily be resolved: Remove the dependency on base because this is already available from elab-util. However, assume we have the following dependency graph where A depends on B and C, and both B and C would depend on D? Would Idris now also shout at us? Could it resolve the issue via the package bounds given in the dependencies of B and C?

I suggest the following: We undo #2469 (for now), try to fix the issue, and then I will run @Z-snails PR version against the current pack collection to see how it goes.

[Z-snails]

Reverting the commit now, I'll make sure to check with pack when I get round to fixing it

[stefan-hoeck]

Transitive dependencies have been (re)implemented in #2584 . Therefore, this can be closed.