Hey, thanks for working on this, #1813 ought to be fixed one way or another.

I don't like this approach because if the user accidentally touches the cabal file, it'll stop updating. That seems very opaque to me, so closing.

@mgsloan but that's why how e.g. make works. User can then touch the package.yaml file? Of course we could save the hash of cabal file contents to be even more sure, but IMHO this is a good first step.

As mentioned, currently stock stack is completely unusable for our company.

Yeah, I considered this further - since you'd have to touch the *.cabal after the modifying package.yaml, this seems fine.

@sol What do you think of making this part of hpack's behavior? Or is it not so necessary because you tend to explicitly run it. It surprised me a bit that hpack writing to the file was based on an equality check rather than timestamp check.

Considering that this also resolves #1828, I think i ought to merge this. Thanks!

One minor reason to do this in hpack is that hpack determines the output cabal file name. This isn't really an issue, though, since we enforce that there's only one cabal file per dir.