Sorry for mass ping -.- I wrongly chose master as the base first.

Having to unmark packages annoys me (as a contributor) more than seeing a broken package fail every time I want to use it.

It's mainly for end users, so they don't have to compile a broken package etc. It can also easily be bypassed in several ways.

I've said it in the past but let this be one more datapoint in the discussion: Having to unmark packages annoys me (as a contributor) more than seeing a broken package fail every time I want to use it.

It’s mostly for the hydra tbh; otherwise it would have to try to build every broken package every time somebody pushes a commit to nixpkgs.

But I agree with @worldofpeace, it’s super easy to wrap pkgs.haskell.lib.markUnbroken around your package, so we can have our cake and eat it, too.

It should be better documented how to mark packages as unbroken at the same time however.

cc @peti @cdepillabout @maralorn

@worldofpeace @Profpatsch I see your points here, especially the one about Hydra resource usage.

Still I want to emphasize that by fixing the problem on my side (e.g. by markUnbroken in my own repo), it becomes less likely for me to contribute upstream. As a low volume contributor I'm often in the situation where I ask myself wether it's worth it to create a completely different patch when the problem is already fixed on my side.

I don't think we can conceptually work around this. We have over 6000 packages from hackage marked as broken. Since there is no way to depublish anything on hackage, a lot of them are probably really of no relevance to anyone.
We need to compromise somewhere, but letting users run without a warning into compiling Haskell packages which then fail at any (possibley) very late point in the process, seems like a no-go for me.

What we should think about is better tooling. I had the idea about a database exposed via a webpage, that shows all haskellPackages and which of them are marked broken. Then in that list we could (if we had the build power) display if it builds despite the broken flag. (Maybe we could make this pull based? Offer a button to trigger a built?) And we could actually offer a button which creates a PR to remove the broken flag. Just toying around here, not sure when I would have the time to implement something like that.