Supported architectures

Distributions for the following architectures are currently supported:

Build	Linking	Supported build hosts
lamdera-[verison]-macos-x86_64	Dynamic	macos-x86_64, macos-arm64 (Rosetta)
lamdera-[verison]-macos-arm64	Dynamic	macos-arm64 + llvm
lamdera-[verison]-linux-x86_64-musl	Static	linux-x86_64/macos-x86_64 (Docker)
lamdera-[verison]-linux-arm64-musl	Static	linux-arm64/macos-arm64 (Docker)
lamdera-[verison]-win-x86_64	Dynamic	win-x86_64

Building an Lamdera compiler binary

Pre-requisites

• ghcup (scripts will attempt to install GHC 9.0.2 + Cabal 3.6.2.0)
• LLVM v13+ (MacOS arm64 only, suggest homebrew brew install llvm@13)

How to install these varies depending on your build host. GHCup is a convenient option for trying out multiple versions.

Once you have these dependencies, running the relevant build-<os>-<aarch>.sh should result in an Lamdera binary.

Macos cross-compiling

Lamdera should compile on both x86_64 (Intel) and arm64 (M-series) chipset macs.

If you have an x86_64 mac, you can only build x86_64 binaries.

If you have an arm64 mac and Rosetta 2, you can also cross-compile the x86_64 binary, however it will require the prerequisite toolchain in the respective CPU flavour, i.e.

• llvm-arm64 + GHC-arm64 + cabal-arm64 => lamdera-macos-arm64
• GHC-x86_64 + cabal-x86_64 => lamdera-macos-x86-64

If you use ghcup, you can force install the x86 tools to replace the arm64 ones like follows:

ghcup install ghc 9.0.2 --force -p x86_64-apple-darwin --set
ghcup install cabal 3.6.2.0 --force -p x86_64-apple-darwin --set
ghcup set ghc 9.0.2

It seems ghcup doesn't currently support multi-arch installs. Check the arch of your currently installed binaries as follows:

file ~/.ghcup/bin/cabal
file ~/.ghcup/ghc/9.0.2/lib/ghc-9.0.2/bin/ghc

Linux builds with Docker

Docker provides us with the convenience of being able to run encapsulated Linux builds on both Linux and MacOS.

The Docker philosophy of immutable layers is great for reproducibility, but a pain for debugging. To get back some of the benefits of tool-level caching, break up expensive operations whenever possible.

I.e. where cabal install run directly will resume based on prior progress (i.e. expensive package compilation), to get similar behaviour via Docker we need to:

COPY lamdera.cabal ./               # only add lamdera.cabal
RUN cabal build --only-dependencies # single layer for building deps based on lamdera.cabal only
COPY ...                            # add all remaining project files afterward
RUN cabal build                     # run the actual elm build

Without this, even changing the readme would mean Docker decides all the packages need recompiling from scratch.