naobservatory/mgs-workflow

SDKMAN! dependency

Closed this issue · 2 comments

mikemc commented

Currently the readme lists SDKMAN! as a dependency, but I think the only real dependency is having a Nextflow-compatible java version installed, which on AWS EC2 I've generally just done through the standard package manager (dnf) and on mac through homebrew, so I'm wondering if it makes to clarify that you don't need to bother with SDKMAN! if you already have java. E.g. perhaps the readme dependencies section could say something like

  1. Java: Nextflow requires java 11 or later (up to 22), which you can install in several ways, including your OS's package manager, homebrew, or the SDKMAN! Java SDK manager (see https://sdkman.io/install or the Nextflow installation instructions)
  2. Nextflow: To install the workflow management framework, follow the installation instructions available here

Alternatively, we could just list Nextflow as a dependency, since I think Java is only needed as a dependency of Nextflow and is covered in the Nextflow instructions, though maybe it's still helpful to have Java explicitly listed.

willbradshaw commented
mikemc commented

Hmm, the current nextflow install instructions that the readme links to say

Nextflow can be used on any POSIX-compatible system (Linux, macOS, etc), and on Windows through WSL. It requires Bash 3.2 (or later) and Java 11 (or later, up to 22) to be installed.

Which I copied into my suggested new text. This is quite broad and is compatible with just installing the most recent java in AWS with dnf.

But the example install they give using SDKMAN explicitly installs "java 17.0.10-tem", suggesting they recommend a specific version. Possibly this ambiguity is a holdover from the change in language you are noting.

It seems to me that if the pipeline requires a specific version of java (such as java 17) to work reliably, then we should add that to the readme, since it sounds like just saying to use SDKMAN on its own won't guarantee any version, and the nextflow docs are at best highly ambiguous on whether it matters if you use java 17 from SDKMAN or any java between 11 to 22.

Perhaps what is most relevant is reporting the specific java you use when you're testing the pipeline. We could tell people the official nextflow req of java 11 to 22, but state that you test the pipeline with a specific version and recommend using that version for debugging or maximum stability