fix: make which available for R (bioconda) post-links in conda v2 builds

[pditommaso]

Problem

Conda containers with R/bioconductor packages (e.g. the nf-core rnaseq TXIMETA_TXIMPORT step, which installs bioconductor-tximeta) build successfully but fail at runtime:

Loading required package: GenomeInfoDb
Error: package or namespace load failed for 'GenomeInfoDb':
  there is no package called 'GenomeInfoDbData'

Root cause

Bioconductor data packages (GenomeInfoDbData, etc.) don't ship their payload in the conda package — a conda post-link script (post-link.sh → installBiocDataPackage.sh) downloads it and runs R CMD INSTALL at install time.

That R CMD INSTALL starts an R session, and bioconda r-base shells out to /usr/bin/which (absolute path) during utils .onLoad — it ignores $PATH and $R_WHICH. The v2 build image mambaorg/micromamba:2-amazon2023 has no /usr/bin/which, so R can't start, the post-link fails, and the data package is never installed. The post-link failure is non-fatal to micromamba install, so the build "succeeds" and produces a broken image.

This is why simply installing conda-forge::which into /opt/conda/bin was not enough — R only looks at /usr/bin/which.

Fix

In the conda micromamba-v2 build stage (as root), symlink the conda which to /usr/bin/which before the user's packages are installed, so bioconda post-link scripts can run R:

FROM {{mamba_image}} AS build
USER root
...
RUN micromamba install -y -n base conda-forge::which \
    && ln -sf "$MAMBA_ROOT_PREFIX/bin/which" /usr/bin/which \
    && (micromamba install -y -n base -f /tmp/conda.yml ...) ...

Applied to all four conda-micromamba-v2 templates (dockerfile + singularityfile, conda-file + conda-packages). The pixi templates also get which as a base package for consistency, though their *-noble build image already provides a system which. The legacy micromamba v1 template is unaffected (its Debian-based image ships which).

Verification

Reproduced and fixed locally with real Docker builds on amazon2023:

• Before: R CMD INSTALL post-link fails (.onLoad failed ... error in running command), GenomeInfoDbData absent, runtime load fails.
• After: post-link succeeds; final image runs library(SummarizedExperiment); library(tximport) → OK.

Unit tests updated: TemplateUtilsTest, ContainerHelperTest (+ CondaHelperTest/PixiHelperTest contains-checks) pass.

Refs: https://community.seqera.io/t/which-is-missing-in-recent-seqera-containers/2718

🤖 Generated with Claude Code