Add baysor-v0.7.1

[khersameesh24]

Added new recipe for Baysor v0.7.1

• baysor is a segmentation method and it used in nf-core pipelines
• pre-compiled binary used to build recipe
• recommended installation method by the developers of the package
• compilation takes a long time (too many dependencies) and usually fails with a timeout, hence the use of binaries

Closes #48883
Closes #64488

[BiocondaBot]

Reposting for @khersameesh24 to enable pings (courtesy of the BiocondaBot):

@bioconda/core

[khersameesh24]

@BiocondaBot , please add label

[ewels]

Not entirely identical to the working prebuilt-binary state of #64488 in 42aeca2, but not far off.

Given that no-one seems to be able to get this to build on bioconda, I don't see any other way for now. Upstream has a branch with some C++ work so hopefully we can do proper builds in the future. LGTM 👍🏻

[mergify]

queue

☑️ Command disallowed due to command restrictions in the Mergify configuration.

Details

• sender-permission >= write

[khersameesh24]

Hi @ewels , it seems like the merge queue is not picking up this PR. Is there something more I need to add? Please let me know if I am missing something here

[quentinblampey]

Hello @khersameesh24,

Thanks for the PR!
I tried it in my nextflow pipeline, but got the following error (more details here):

InitError(mod=:Makie, error=Base.IOError(msg="mkdir("/opt/conda/lib/baysor/share/julia/scratchspaces/ee78f7c6-11fb-53f2-987a-cfe4a2b5a57a/makie"; mode=0o777): permission denied (EACCES)", code=-13))

I tried several things, but couldn't make it work. Also, when using the image with default Docker params locally, it works great, so it's likely related to nextflow env variables or the containerOptions.

(I don't know yet if it's specific to my pipeline or if the cause is this recipe)

[quentinblampey]

I tried several fixes in my pipeline, but couldn't make it work. I really suspect that the fix should come upstream (i.e., in this bioconda recipe).

It seems Julia/Makie is trying to create scratch/cache directories within the conda install prefix (/opt/conda/...), which is apparently not writable in nf-core CI’s non-root container.
I'm not really familiar with Julia, but both ChatGPT and Gemini agree on setting JULIA_DEPOT_PATH (and optionally XDG_CACHE_HOME) to ${TMPDIR}//tmp, rather than patching my Nextflow pipeline.

E.g., replacing the wrapper in the build by:

cat > "${PREFIX}/bin/baysor" << 'EOF'
#!/bin/bash

export LD_LIBRARY_PATH="${CONDA_PREFIX}/lib/baysor/lib:${LD_LIBRARY_PATH}"

export TMPDIR="${TMPDIR:-/tmp}"

export JULIA_DEPOT_PATH="${TMPDIR}/julia-depot"
export XDG_CACHE_HOME="${TMPDIR}/xdg-cache"

mkdir -p "${JULIA_DEPOT_PATH}"
mkdir -p "${XDG_CACHE_HOME}"

exec "${CONDA_PREFIX}/lib/baysor/bin/baysor" "$@"
EOF

WDYT @ewels @khersameesh24?

[ewels]

Sounds sensible to me. We've had similar stuff in nf-core pipelines before, but equally having it in the bioconda recipe would make the package functional for all users, which seems like a net win. So worth a PR 👍🏻

[khersameesh24]

Hi @quentinblampey @ewels
A PR with the fix has been raised

[quentinblampey]

Hi @khersameesh24, unfortunately, I still get the same issue with the new recipe build...
Did you find a way to make it work for your nextflow pipeline?

[khersameesh24]

Hi @khersameesh24, unfortunately, I still get the same issue with the new recipe build...
Did you find a way to make it work for your nextflow pipeline?

Hi @quentinblampey , unfortunately the image generated with the earlier build is cached on seqera containers. If you built your image again, there is a high chance it might be returning you the cached image built around 11th May. Did you use wave containers or did you build the image with a Dockerfile?

Our temporary fix is to use
quay.io/khersameesh24/baysor-0.71

[khersameesh24]

Hi @khersameesh24, unfortunately, I still get the same issue with the new recipe build...
Did you find a way to make it work for your nextflow pipeline?

Hi @quentinblampey , unfortunately the image generated with the earlier build is cached on seqera containers. If you built your image again, there is a high chance it might be returning you the cached image built around 11th May. Did you use wave containers or did you build the image with a Dockerfile?

Our temporary fix is to use `quay.io/khersameesh24/baysor-0.7

These are some more details
Baysor Container

the conda lock file currently has baysor-0.7.1-hab16a5f_0 but should have baysor-0.7.1-hab16a5f_1

[quentinblampey]

Oh I see, it may be the issue indeed! The thing is, I need to create a container with both sopa and baysor, so I can't use the image you mentioned...
The image I used was created today via the wave CLI (with baysor + the latest sopa version). Do you know how to force using the new baysor build?

[khersameesh24]

Oh I see, it may be the issue indeed! The thing is, I need to create a container with both sopa and baysor, so I can't use the image you mentioned... The image I used was created today via the wave CLI (with baysor + the latest sopa version). Do you know how to force using the new baysor build?

Unfortunately, this is not supported right now on seqera containers, but I have started a discussion on slack channel and @ewels is aware of this, we have similar problems with other tools available via bioconda as well. Can you manually change your conda lock file and build an image? Maybe ask the core team to put it on quay.io as a temporary measure?

[quentinblampey]

Thanks for your help, I'll try that 😊