# cabal-flatpak: Generate a FlatPak manifest from a Cabal package description

[ bsd3, distribution, program ] [ Propose Tags ]

Generate a FlatPak manifest from a Cabal package description.

Inspired by Richard Szibele's stackpak: https://gitlab.com/rszibele/stackpak#readme

#### Maintainer's Corner

For package maintainers and hackage trustees

Candidates

Versions [RSS] 0.0, 0.1, 0.1.0.1, 0.1.0.2, 0.1.0.3 Changes.md aeson (>=1.2 && <1.5), aeson-pretty (>=0.8.7 && <0.9), base (>=4.5 && <5), bytestring (>=0.9 && <0.11), cabal-plan (>=0.5 && <0.6), containers (>=0.4.2 && <0.7), http-client (>=0.5.12 && <0.7), http-client-tls (>=0.3.5.3 && <0.4), http-types (>=0.10 && <0.13), optparse-applicative (>=0.11 && <0.16), pathtype (>=0.8 && <0.9), shell-utility (>=0.0 && <0.1), text (>=1.2 && <1.3), utility-ht (>=0.0.12 && <0.1) [details] BSD-3-Clause Henning Thielemann Henning Thielemann Distribution http://hub.darcs.net/thielema/cabal-flatpak/ this: darcs get http://hub.darcs.net/thielema/cabal-flatpak/ --tag 0.0head: darcs get http://hub.darcs.net/thielema/cabal-flatpak/ by HenningThielemann at 2019-09-21T10:19:03Z LTSHaskell:0.1.0.3, NixOS:0.1.0.3, Stackage:0.1.0.3 cabal-flatpak 1256 total (6 in the last 30 days) 2.0 (votes: 1) [estimated by Bayesian average] λ λ λ Docs not available Last success reported on 2019-09-21

[back to package description]

## Scope

This program generates a large Flatpak manifest in JSON format with the versions of all transitively imported Haskell packages. This is required by Flathub since this platform downloads all source files for you and lets you build only in a sandbox without access to the internet. The detailed version information can also be useful for you for reproducible builds. However, for maintaining your own Flatpak repository or for distributing Flatpak bundles it is not strictly necessary to have such a verbose manifest. You may get results quicker by writing you own manifest which simply runs cabal-install or stack. stack should give you reproducible builds but its package database contains only a subset of Hackage. cabal-install gives you access to more Haskell packages, but the build may fail at a later point due to lax version bounds in dependent packages.

## Usage

### Create a Flatpak manifest

The usage is a bit cumbersome, because for Flathub you may only build manifests for released versions of your package, whereas you certainly want to manage manifests in the working copy of your repository.

First let cabal build a plan.json for your released package:

$cd /tmp$ cabal unpack my-package-0.1.2
$cd my-package-0.1.2$ cabal new-build --dry-run --disable-tests --disable-benchmarks


cabal-flatpak reads plan.json via the cabal-plan library.

Now create a flatpak.cabal.json file in your working copy. It contains a custom JSON object with information needed by cabal-flatpak and also a template for the generated manifest. For an example see the configuration file for cabal-flatpak.

$cd /path/to/my-package$ cabal-flatpak --directory=/tmp/my-package-0.1.2 flatpak.cabal.json flatpak.json


The --arch option allows you to put build information for one or more architectures into one manifest. Usually, only ghc and cabal-install binaries depend on the architecture and the whole lot of Haskell packages can be build with the same commands.

There are two build modes: One builds all modules individually using plain Cabal, the other one builds all modules in one go using cabal-install. You can enable the second mode using the --cabal-install option. These are the differences:

• Cabal mode needs less dependencies. cabal-install needs a pre-built binary for your architecture.

• Cabal mode can pass options to the build of specific packages. cabal-install can only pass options to all packages at once. If -fbuildExamples means something different to different Haskell packages, then this will fail. Even in cabal-install mode we need to preprocess each package and in this stage we could alter flag switches by patching Cabal files. We could also build packages with requested flags in separate stages. Currently, we don't try any of these strategies.

• cabal-install enables parallel builds. This builds significantly faster. Cabal mode on the other hand allows Flatpak to cache build results of individual packages. This can accelerate re-builds. However, Flatpak does not know the dependency graph and thus simply rebuilds anything after a module that must be rebuilt.

Unfortunately, you cannot simply switch between the two modes simply by adding or removing the --cabal-install option. The reason is that you must maintain the data for the main package manually in the "main-sources" object and Cabal requires the source tarball unpacked whereas cabal-install needs it packed. Thus Cabal needs "type": "archive" and cabal-install needs "type": "file".

For some packages you need to build dependencies on external C packages. cabal-flatpak cannot generate according build instructions for you. However, you can re-use build instructions you found useful. Flatpak-builder supports this itself. In the "modules" list you cannot only put module JSON objects, but also plain strings. Such a string is interpreted as path to a separate file containing a Flatpak module JSON object. I add such JSON files to the FFI packages I maintain.

### Build the Flatpak package

You may refer to the Makefile that is shipped with cabal-flatpak for how to eventually build the Flatpak package. The command line is:

$flatpak-builder --force-clean --repo=$FLATPAK/repository --state-dir=$FLATPAK/builder/$FLATPAK/build/my-package $<  Flatpak consumes pretty much storage thus I set$FLATPAK to a directory on a separate harddisk partition. The --repo option points to your Flatpak repository. This is where flatpak-builder stores the compiled package. The Flatpak repository contains all versions of all your Flatpak packages. If another user has access to it, she can easily install and update Flatpak packages. The --state-dir option points to a directory that caches all downloads and build artifacts of Flatpak. You may share it between different projects. The DIRECTORY argument names the path to where Flatpak builds your project. It may not be shared between projects, but you can safely delete it after a build and the --force-clean option triggers exactly this when you re-build your project. The directories specified by --state-dir and DIRECTORY must reside on the same file system.

You may also extract a single package file for a certain version of your package from the repository. This can be handy for a one-time install but disallows the user to easily get updates of your program.

$flatpak build-bundle$FLATPAK/repository \
my-package-0.1.2.flatpak com.my_domain.my-package \
--runtime-repo=https://flathub.org/repo/flathub.flatpakrepo


## Known issues

### Cabal revisions

Cabal package description files can be updated on Hackage without altering the package version. Every update increases the Cabal file revision. Unfortunately, plan.json does not contain the revision number, it just refers to the currently most recent Cabal file for each Haskell package. This is a fundamental problem since Flatpak builds must be reproducible.

We have currently no solution for that problem. If someone updates the Cabal file of one of your dependencies, then the Flatpak build will break with an error due to a SHA256 hash mismatch. In this case you should find out the original revision and add that to the Hackage path in the Flatpak manifest. E.g. if the path to a Cabal file is say https://hackage.haskell.org/package/cabal-flatpak-0.0/cabal-flatpak.cabal then the path to revision 1 is: https://hackage.haskell.org/package/cabal-flatpak-0.0/revision/1.cabal.

We could try to solve the problem by downloading Cabal file revisions from Hackage for every imported package until we find the one with a matching SHA256 hash. We could try to do this offline and scan Cabal's local package database using a program like: https://github.com/haskell-hvr/hackage-index . None of these ideas is implemented, yet.