add linux-aarch64 built

[stschiff]

As requested on Bioconda here, we would like to add a linux-aarch64 built.

[stschiff]

Just an update for @dslarm, who requested this: I have managed to make Arm builds here on GitHub, but in the meantime the Docker-build for CentOS7 failed, so porting to bioconda is still not finished.

[dslarm]

Hi @stschiff - thanks for working on this and for the update - great progress!

[stschiff]

This is done now. The latest release has a linux-arm executable. I have not yet managed to bring back the linux-centos version that used to be needed to bioconda. I need to check whether that's still the case or whether we can just use the current executables.

[stschiff]

OK, so I have some definite news on this, and I think I will also post this somewhere on bioconda. Let me unpack the issues in order:

Issue 1) Compiling Haskell programs on bioconda is extremely hard. I've spent an entire week some years back, and I did not succeed. The problem is mostly caused by the vintage CentOS the bioconda build environment runs on.
-> Solution 1) Build Haskell tools on GitHub actions and load the pre-built executables within bioconda.

Issue 2) The pre-built executables from GitHub for Mac-Intel architectures works just fine. But for Linux, the issue is that the github-runner on which the exec is built uses a recent Ubuntu, with relatively recent versions of GLIBC. These are not available on the vintage bioconda-build system, so the executables built on GitHub do not run on bioconda.
-> Solution 2) Use a custom docker container on GitHub to emulate the bioconda build environment.

Issue 3) Solution 2 works for Linux Intel Architecture. But for the AARC64 architecture we face another problem: The Haskell compiler GHC-9.6.6 cannot be compiled without relatively recent versions of GLIBC. So I simply cannot compile the Haskell-compiler on the build system required by bioconda.

Conclusion: The specific build-environment on the bioconda Linux-AARC64 environment precludes Haskell builds.

So, sorry @dslarm ... but aarch64 Haskell builds on bioconda will not easily become a reality, I'm afraid.

[dslarm]

Thanks for your efforts here, and the write up - it seems you're not the first to encounter the glibc version issue. Within bioconda, bioconda/bioconda-recipes#46186 charts the issue.

Within conda-forge, there seem to be some mitigations possible as CUDA specified higher minimum glibc versions than prevalent at some point in the past (https://conda-forge.org/news/2024/11/22/new-images/) - although I don't seem to be able to make those tricks work here.

I don't see any mention of bioconda changing base image any time soon either :(