Issues with modules that require binaries (wrappers)

[bmpvieira]

@yannickwurm: "Some major work is required for modules which require binaries. Currently using samtools (etc) involves download & compiling... this is extremely prone to fail & a big maintenance effort. Instead, rely on local docker install (so each module relies on a single docker file. (thus each module would include a dockerfile)."

So, either we:

• Rely on Docker like Yannick says (issue: Docker becomes a dependency);
• Don't provide at all the binaries (just the wrapper) and push the responsibility of installing to the user (sucks and hard to control versioning);
• Try to provide pre-compiled binaries (but then, we're doing package management...). @maxogden might have some ideas;
• Turn some of those tools into native addons (hard to do and maintain in sync with upstream).

[doomedramen]

for ncurses dependencies, in debian:
sudo apt-get install libncurses5-dev
or if that fails
sudo apt-get install ncurses-devel ncurses

[jamidwyer]

I don't know the best solution for installing on Windows, but in case anyone is stuck, removing bionode-sam from bionode's package.json before installing worked.

[yannickwurm]

combine 1 and 2:

• Don't provide at all the binaries (just the wrapper) and push the responsibility of installing to the user

but in case they have a docker install, provide a dockerfile that includes the versions you've explicitly tested on. (not a dependency but an easy option)

[bmpvieira]

Another project that is wrapping bioinformatic tools is the Common Workflow Language, and after chatting with @tetron at #biohack15 he added to the current implementation of the specification some support for piping in/out NDJSON. This is the main requirement for using CWL with bionode pipelines. However, it's still an early project and there's very few tools wrapped (mostly just examples).

As for getting bioinformatics tools deployed/installed, Docker is great but still has issues (e.g., security), and there's a lot of alternatives popping up (e.g., vagga, hyperos.io, Rocket) so we probably don't want to commit yet to one solution. Plus, this might be out of scope for bionode in any case (we don't want to do package management...)

Alternatives to containers are package managers like Homebrew linux (lots of bioinformatics tools, but some won't install depending on host machine toolchain) and GUIX (functional package manager that is completely isolated and independent of root filesystem, plus containers support is coming for extra isolation).