Skip to content

Latest commit

 

History

History
110 lines (87 loc) · 3.98 KB

README.md

File metadata and controls

110 lines (87 loc) · 3.98 KB
Raw

##Setup

These scripts help create recipes for bioconductor packages. You can set up an environment for running them using:

conda config --add channels bioconda
conda create -n bioconductor-recipes --file requirements.txt
source activate bioconductor-recipes

Single Bioconductor recipe

To make a single recipe:

./bioconductor_skeleton.py --recipes ../../recipes GenomicRanges

Note that the case of the bioconductor package must match that of the actual R package. Use --force to overwrite the recipe. The conda recipe ../../recipes/bioconductor-genomicranges will be created.

This script:

  • parses the Bioconductor web page for the current version
  • checks to see if a tarball for the version is available on bioaRchive and if so uses that for long-term stability. Otherwise it uses the tarball from the Bioconductor page.
  • downloads the tarball to the cached_bioconductor_tarballs dir and extracts the DESCRIPTION file
  • parses the DESCRIPTION file to identify dependencies
  • converts dependencies to package names more friendly to conda, specifically prefixing bioconductor- or r- as needed and using lowercase package names
  • bumps the build number if package versions are the same but something else in the recipe changed
  • calculates md5sum on the cached tarball
  • writes a meta.yaml and build.sh file to recipes/<new package name>.

Like building recipes with conda skeleton, it may take a few rounds of trying to build the package, failing because of missing dependencies, running bioconductor_scraper.py again to build dependency recipes, and trying to build again:

(
    cd ../../
    docker pull bioconda/bioconda-builder
    docker run \
        -v `pwd`:/bioconda-recipes bioconda/bioconda-builder \
        -v --packages bioconductor-genomicranges
)

Updating recipes

This is most needed when there is a new release of Bioconductor.

When possible, the tarball URLs in the recipes created by bioconductor_skeleton.py point to the bioaRchive tarballs, which should have more long-term stability. However, sometimes the latest package version listed on the bioconductor page at the time of recipe creation is not available in bioaRchive. In this case the recipe points to the bioconductor site's tarball. However, a new release of Bioconductor will break that URL.

Two helper scripts make this as painless as possible:

update_bioconductor_packages.py does the following:

  • searches the recipe dir for bioconductor-* and r-* packages
  • returns them topologically sorted such that if the packages are built in order, dependencies will be built first.

update-bioconductor-packages-wrapper.sh does the following:

  • rolls back all changes in the recipes dir for any bioconductor- or r- recipes
  • calls update_bioconductor_packages.py to get the toposorted list of existing recipes to work on
  • for each R package, runs conda skeleton cran --update-outdated on the recipe
  • for bioconductor packages, runs bioconductor_skeleton.py with --force to overwrite the existing recipes. This script also pays attention to whether or not there is a new bioaRchive tarball (even if the package version was unchanged), in which case it bumps the build number to trigger a re-build of the recipe

So in practice, when a new Bioconductor release comes out, you should simply have to do:

./update-bioconductor-packages-wrapper.sh ../../recipes
(
    cd ../../
    docker run \
        -v `pwd`:/bioconda-recipes bioconda/bioconda-builder \
        -v
)

Note that some packages specify dependency versions using strings that are not allowed by conda. For example, GenomicFeatures 1.22.6 specifies RSQLite 0.8-1. That -1 is not allowed by conda. Rather than be tricky about automatically fixing version strings, currently you have to fix this manually. In this case, since the earliest RSQLite conda package is 1.0.0, we can either get rid of the version specification or change it to 1.0.0.