If anybody else is interested.
For now, I settled with a Stack based solution.

`STACK_ROOT=<some-temp-folder> stack build --dry-run --prefetch`

works like a charm. What I didn't tackle yet is having multiple ghc installations in that folder. Somehow I will find a solution for that as well.

For now I live with the fact that

`STACK_ROOT=<extracted-archive-from-prior-step> stack build --install-ghc`

will connect to the internet to download ghc.

Thanks for all the responses. I'll send an update when the ghc thing is solved as well.

Best
Jan

Herbert Valerio Riedel <hvriedel@gmail.com> schrieb am Do., 24. Nov. 2016 um 22:48 Uhr:
Hi,

On 2016-11-24 at 07:57:57 +0100, Jan von Löwenstein wrote:
> Consider the following use case:
> I have to ship a Haskell application as source, bundled with its
> dependencies and a script that can produce the binary on a machine without
> internet connectivity.

This situation is not that uncommon; one example is when you have
build-bots which a more or less cut off from the internet (such as
Ubuntu's launchpad buildfarm)

> Is that possible with Stack or Cabal?
> Any pointers would be much appreciated.

> Ideally I would like to build each dependency package individually. That
> way I could cache results per Haskell package and don't need to rebuild
> dependencies until they actually change.

This sounds like something I'd use cabal's filesystem-local repository
feature[1] for, and it should be possible to specify such local
repositories in your `cabal.config` or `cabal.project`[2] files to
produce a self-contained source distribution.

If you're interested in pursuing this approach or have any questions
about it, let me know!


 [1]: http://cabal.readthedocs.io/en/latest/installing-packages.html#repository-specification
 [2]: http://cabal.readthedocs.io/en/latest/nix-local-build-overview.html

Cheers