Insane library upgrade ordering

[l29ah]

haskell-updater doesn't produce a proper order for the packages to be built in, and portage can't figure it out (as it thinks that all the libraries-to-be-rebuilt are present and in a good condition?), resorting to random-ish order and predictably failing at the first package due to the broken deps after ghc upgrade.

[solpeth]

Yes, it is a pain. But I only experience failures with FEATURES=test enabled. Is this your case? If not that's something new to me.

[l29ah]

No, tests aren't enabled.

[solpeth]

Okay. I'm out of my depth with haskell-updater so hopefully someone else can take a look.

[l29ah]

Seems like haskell-updater -- --with-bdeps=y --ignore-built-slot-operator-deps=y does the trick.

[solpeth]

Good to know! Thanks for the tip!

[trofi]

That is usually a sign of circular dependencies (explicit or implicit). It should not happen for either USE=test or USE=-test. Both types of circular dependencies should be tracked down and cycles broken if possible. We don't have nice tools to do that though.

I usually brute-force my ways into the cycles like that:

1. pick first package that fails to build
2. try to rebuild it's depends, check what breaks to build
3. collect graph manually (usually it's not too many steps)

Once done spot the cycle.