Building conda packages for CTA

The following information is for CTA pipelines code administrators to create packages (it is not needed by regular users)

Create a new recipe for your revision

For each release, there should be a subdirectory in this repo called <package>-<version>.conda/ that contains the recipe. This way we can build past versions as well in an automated fashion.
The <version> should be major.minor or major.minor.sub, for example: ctapipe-0.2.0.conda/. For re-builds of the same version release you do not need to create a new directory (just increment the internal build number).

Create the recipe for your release

1. create a subdir for your release as above.

2. Create or select a version branch/tag of the package (e.g. v1.2.0)

3. edit <package>.conda/meta.yaml (copy it from previous version):

   • change the git_rev field to be the branch/tag name you chose
   • make sure the entry_points are the same as in the setup.py file of the main package, otherwise any command-line utils will not be installed
   • make sure the requirements/build list is also the same as in setup.py (any new dependencies should be added)

Build the packages locally

Run the following to create the package files. The output will be usually in <ANACONDA>/conda-bld/<ARCH>/*.bz2, where <ANACONDA> is the path to your Anaconda or miniconda installation, but you can check for sure by running conda build ctapipe.conda --output, which will give you the path to the package.

conda build [--python=<version>] <package name>.conda

# for example:
conda build --python=3.5 ctapipe-0.3.3.conda

Next, make a new clean conda environment to test that everything works and all the dependencies are ok:

conda create -n ctatest 
source activate ctatest 

Finally, try installing the new ctapipe package into this environment. You should get pyhessio and other dependencies automatically. For now you must use the --use-local option, until the package is uploaded to Anaconda Cloud in the cta-observatory channel (see next step).

conda install --use-local ctapipe=<version>  # leave out =version to get latest
ctapipe-info --version --dependencies --tools

If all goes well, and your account has upload access for the cta-observatory organization's channel on Anaconda Cloud, you can upload the packages like this:

anaconda login --username <your anaconda username>
anaconda upload --username cta-observatory <package files>

From then on, users can install the package via:

conda install -c cta-observatory <package>

Notes

• You must build each package on a 64-bit macOS system and a linux system to ensure compatiblity with all CTA machines. However, if the package has no compiled C code in it, you can use conda convert <package filename> -p <target> to convert it automatically. Targets are osx-64 or linux-64.

• For Linux, please only build on a SL6.X or compatible system (SL6.8 is used currently) to ensure the package works on CTA grid machines. Building on newer Linux distros will cause problems due to changes in glibc that are not backward compatible.

• To build for multiple python versions (e.g. 3.4, 3.5, 3.6), you need to re-run the conda build command with for example the --python=3.5 option, once per python version, and upload all packages