Hi Bjoern, That looks much better indeed ;) The only problem I still have then is that I need R 3.1.0 for a bioconductor 2.14 package (have send a new mailing list msg for that). Looking at the xml of other versions it's not something I will easily do myself. What will happen if I do not specify the R dependency ("package_r_3_0_3" in your example code) but do specify the download/install of packages, guess these get installed in the "default" R instance? Related to that, how can I call a specific instance of R in de tool.xml without specifying the full path to the tool. Eg, in the tool.xml I now do: <command> /path/to/lib/R/R-3.1.0/bin/Rscript /path/to/galaxy-dist/tools/testdir/tool.R $config </command> Where normally you can do: <command interpreter="Rscript"> tool.R $config </command> Thanks again! Stef ----- Original message ----- From: Björn Grüning <bjoern.gruening@gmail.com> To: Stef van Lieshout <stefvanlieshout@fastmail.fm>, galaxy-dev@lists.bx.psu.edu Subject: Re: [galaxy-dev] R bioconductor dependencies when creating toolshed installation Date: Tue, 17 Jun 2014 15:17:36 +0200 Hi Stef, for R packages we have a special installation routine that will (hapefully) make your life easier.
I'm running into some difficulties on how to setup the installation procedure for a galaxy tool which executes an R script and has certain dependencies (mainly bioconductor packages). R can deal with dependencies, packages can be installed with install.packages (has a "dependencies" argument) or biocLite() for bioconductor packages.
Yet, now I want my tool to be available at toolsheds. To do this I see several options:
Great!
1) setting up tool_dependencies.xml with "R CMD INSTALL" for all packages. BUT: need to download all dependencies before install, and can older versions still be downloaded? Maybe need to upload them to toolshed too..
It is all a matter of how reproducible you want to have your tool. If you want 100% reproducibility, you need to mirror the source packages somehow, because bioc will not store older versions. At least that is not guaranteed. I'm using a special github repository for that purpose: https://github.com/bgruening/download_store R CMD INSTALL is not needed, see below.
2) setting up tool_dependencies.xml to call an installation script with Rscript (where I could use install.packages), BUT: Dependencies are taken care of. But how do I select specific (older) versions, because if I dont, installing at different time can give different version.
Older versions is not possible as far as I know.
3) creating a repository for each package and have all of them as requirement in my galaxy tool. BUT: a lot of work for a lot of dependencies
Imho, we should have one R repository with a handful of standard packages included in the toolshed. Like packages_r_3_0_1. You should depend on that repository and additionally define one second dependency. Lets say your tool is called deseq2 than create one additional tool_dependencies.xml file called package_deseq2_1_2_10. In that definition you will install every dependency you need in addition to R. Here is one example: https://github.com/bgruening/galaxytools/blob/master/deseq2/tool_dependencie... https://github.com/bgruening/galaxytools/blob/master/orphan_tool_dependencie... The really nice part is the "setup_r_environment" function from the toolshed. It will install source packages for you automatically. All you need to do is to name the package or, as shown in the example, specify the location of the source package. The only downside is that the order of these packages is important. If you are interested we have a script that will give you the correct dependency tree of a given package. Hope that helps, Bjoern
All have pros and cons, how do people deal with this?
Stef ___________________________________________________________ Please keep all replies on the list by using "reply all" in your mail client. To manage your subscriptions to this and other Galaxy lists, please use the interface at: http://lists.bx.psu.edu/
To search Galaxy mailing lists use the unified search at: http://galaxyproject.org/search/mailinglists/