Skip to content
/ condev Public

Creates conda development environments

License

Notifications You must be signed in to change notification settings

maddenp/condev

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

57 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

condev

ci

Creates conda development shells and provides related convenience tooling.

Getting Started

The following subsection provides instructions on obtaining, installing, and activating conda. The two remaining subsections describe two methods for installing condev and its required tools into your conda installation: 1. Using a prebuilt package, or 2. Bootstrapping.

Installing conda

conda itself may be provided by a Miniconda, Miniforge, Mambaforge, or Anaconda installation. Prefer one of the first three for a lightweight installation providing only what you need. Miniconda is the official distribution from Anaconda, Inc., and defaults to using their package collection. Miniforge is equivalent to Miniconda except that it defaults to using the conda-forge package collection: A community-curated collection of high-quality packages for which all recipes are available for public inspection. Mambaforge is identical to Miniforge except that includes the mamba tools. If you are unsure, use Miniforge, as shown below.

wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
bash Miniforge3-Linux-x86_64.sh -bfp /desired/path/to/conda
source /desired/path/to/conda/etc/profile.d/conda.sh
conda activate

Using a Prebuilt Package

With your conda activated, the following command install the latest available condev package:

conda install -y -c maddenp condev

This also installs dependency packages providing:

  • conda-build and conda-verify for building conda packages and interpreting their metadata
  • jq for extracting select metadata values from conda-package metadata
  • A late-model make for using the convenient targets defined by the condev Makefiles

You can also search the maddenp channel for available versions with conda search -c maddenp --override-channels condev and install a specific version by replacing condev with condev=<version>[=build] in the preceding conda install command.

Bootstrapping

With your conda activated, the following steps install required dependency packages, builds the condev conda package, installs that package into the base environment, then verifies that the expected condev programs are available:

conda install -y conda-build conda-verify jq make
make package
conda install -y -c local condev=$(jq -r .version recipe/meta.json)=$(jq -r .build recipe/meta.json)
which condev-meta condev-shell

After executing this procedure, you should have a condev-equipped conda installation that you can activate in the future with the command source /path/to/condev/conda/etc/profile.d/conda.sh && conda activate.

General Use

If you adhere to a few conventions in the layout and configuration of your project, you can use your condev-equipped conda installation to:

  • Obtain a development shell in which you can edit and test your code interactively, in an isolated environment with all dependency libraries available, with make devshell
  • Build a conda package based on your project's recipe with the make package command
  • Create a conda environment based on your project's recipe with the make env command
  • Execute your project's defined code-quality tests with the make test command (or the more granular make lint, make typecheck, and make unittest commands)
  • Auto-format your Python code with the make format command

The Bash shell created by make devshell may start with only a subset of the host shell's environment. If a .condevrc file exists in your home directory, it will be sourced to do any environment setup you require. For example, you might create a ~/.condevrc file with the content source ~/.bashrc.

Demo

The demo subdirectory of this repo offers a prototype project (i.e. files/directories that would appear in the root of a separate git repository) that leverages condev to demonstrate the use of the various convenience make targets. It also demonstrates one way in which hybrid Python/C projects might be handled.

make devshell

Use make devshell to create and activate a conda environment in which all dependency packages are available, and package's Python code is live-linked for a fast edit/test loop. After executing the the Getting Started procedure above, try the following:

cd demo
make devshell

This will create a conda environment with your project's build, host, run, and test packages all installed and ready to use. The Python code under src/hello is live-linked into the environment via a setuptools editable install, so any changes you make can be immediately executed, tested, etc. You will see your Bash prompt prefixed with the name of the development environment, (DEV-hello).

If this were a pure-Python codebase, you could now run the heythere entry-point console script (defined by src/setup.py). Try that and see how it fails. Since this project's Python code relies on a C function (which also lives in the project, so cannot be satisfied by installing a dependency package), that needs to be build first:

(cd src/world && make install)

Now run heythere and it should succeed.

The demo project provides auto-formatting via black and isort. Run make format to auto-format all Python code in the project. This should do nothing if you haven't changed the code since you cloned the repo, but feel free to edit e.g. src/hello/core.py, then run make format to see the tools in action.

You can run code-quality tools -- linter, type checker, and unit tests with a coverage report -- by running make test. These tests rely on pylint, mypy, pytest, and coverage, primarily configured via the pyproject.toml file in the project's src/ directory. You can also run tools individually with make lint, make typecheck, and make unittest. Again, these tools should all pass if you have not changed the code (provided you have build the C code as described above), but you can tweak the code to see how the tools respond to code-quality issues.

When you are finished with development work, you can simply exit the development shell with exit or CTRL-D. If you later type make devshell again, the existing environment will be activated nearly instantaneously.

NB: As a rule of thumb, you should manually remove your development environment (e.g. conda env remove -n DEV-demo) if you change the contents of recipe/, and especially if you change the meta.yaml file, which defines required dependency packages. Likewise, the file recipe/meta.json is generated by several make targets, but only if certain other files under recipe/ change. There could be other circumstances under which this file should be regenerated, however, so manually remove it and re-run your make command if in doubt.

Run a single command in a development shell

To support automation scenarios, you may export environment variable CONDEV_SHELL_CMD and then run make devshell (or its underlying condev-shell script directly), which will result in the variable's value being executed as a command in the devlopment shell, followed by the development shell exiting and returning control to the caller.

make package

While not in a development shell (exit from DEV-hello if necessary), run make package to invoke conda-build to build the demo code as a conda package. (Note that the same tests run by make test will be run by conda-build, and must succeed for the final package to be built -- so to save time, ensure that make test in a development shell succeeds before attempting make package.) When the package has been built, you can verify its availability with

conda search -c local --override-channels hello

Depending on your needs, you could upload this package to a public Anaconda channel, to a private conda-package server, or simply access it from local disk, creating conda environments that include it.

make env

Use make env to create a conda package based on the current directory's recipe, then create a conda environment based on that package -- pulling in any run-time dependency packages it declares in its recipe/meta.yaml file -- named based on the package name, version number, and build number. Try:

make env
version=$(jq -r .version recipe/meta.json)
buildnum=$(jq -r .buildnum recipe/meta.json)
conda activate hello-$version-$buildnum
heythere

Conventions and Advice

The demo project demonstrates a number of conventions (or in some case requirements) you may find helpful when working with condev and/or with conda and/or with Python projects in general:

  • recipe/build.sh is a standard conda-build file and is executed to build your code and install it into the conda-build-supplied $PREFIX tree for packaging. Simple build recipes can be provided directly in meta.yaml as an alternative, but build.sh allows maximum flexibility.
  • recipe/channels specifies a list of channels conda is allowed to use to obtain packages. In calls to conda create, conda install, etc., these channels are translated into -c (aka --channel) command-line options. Channels are consulted in priority order; packages satisfying requirements are taken from the highest-priority channel. The local channel (i.e. packages built by and contained in your local conda installation) is always implicitly added as the lowest-priority channel. This file is required by the condev tools, but is not a standard file from the perspective of conda-build.
  • recipe/conda_build_config.yaml specifies so-called build variants. conda-build will generate the Cartesian product of all elements in this file, building a separate package for each combination.
  • recipe/meta.yaml is well described.
  • recipe/run_test.sh is a standard conda-build file, is executed late in the package-build process, and must run successfully for the build to complete. Normally, the tests executed in this script are meant to test the built and packaged code in its run-time environment; however, the demo project, by adding appropriate entries to the test source_files section of meta.yaml, also arranges for everything available in a development shell to also be available at conda-build test time, such that the same code-quality tests executed by make test at development time can be exercised one last time before the final package is built. (In fact, make test works by executing run_test.sh in the development shell, to avoid build-time surprises.) It would be appropriate to extend these tests to perform integration tests with realistic data (perhaps by creating test-data conda packages and having them installed via the test requires section in meta.yaml) that are not part of the development-environment code-quality tests, but this is not done in the demo project.
  • src/pyproject.toml specifies most options for the code-quality tools listed in the test requires section of meta.yaml and exercised by make test / recipe/run_test.sh. Modify it to your liking.
  • src/setup.py is used both to install Python code (and potentially non-Python scripts, as shown in the demo project) at package build time, and to create an "editable install" into the development shell. It relies on metadata extracted by the project's conda recipe by the condev-meta tool, which is then written to recipe/meta.json. The project Makefile ensures that, except for in some rare cases, meta.json is regenerated as needed. So, prefer to rely on the provided make targets when possible: The condev-shell program expects to be run in a directory where src/setup.py is available, and running make devshell (or make -C /path/to/repo/root devshell) ensures that, if src/setup.py exists, it will be found. If you need to call condev-shell directly, do so from the repo root if you have a src/setup.py.
  • src/setup.py also defines so-called entry-point console scripts: The command-line programs that your package provides users. Prefer writing your Python code as a library (i.e. do not include #! at the top of, or if __name__ == "__main__" at the bottom of, your .py files) and rely on setuptools (via setup.py) to create and install entry-point scripts that load and call the appropriate package, module, and function.
  • Use importlib.resources to provide configuration files, etc., to your code. Python will always know where to find resources so configured, regardless of whether your code is running in a development shell or a final, deployed conda environment, eliminating any headaches about paths.
  • If these conventions are followed, demo/Makefile should be usable for most any Python project.

Notes

  • There is a lot about conda-build that all this glosses over. It is a powerful tool for building, packaging, and versioning code from nearly any language, and covers countless complex use cases. So, the the documentation is invaluable. The Defining metadata (meta.yaml) page is especially handy. There's no way to avoid complexity in some cases, though using these condev tools and their associated conventions can often hide a lot of it.
  • You might like to create a single conda (e.g. Miniforge) installation that multiple users of a multi-tenant system can share. Beware: Many conda packages contain files that are group-writable, and if your shared conda installation is chgrped to a group that your users are also members of, they may be able to, even if by accident, overwrite files that should be read-only to them. To work around this: 1. Make sure that the conda installation's files are chgrped to a group that only contains the appropriate user(s); and 2. Have each user export CONDA_PKGS_DIR=$HOME/.conda/pkgs or similar. The first step ensures that arbitrary users cannot modify any part of the conda installation but, by itself, would also lead to errors when users ran commands like conda create, as they would not have permissions to download package files into the conda directory, as conda requires. The second steps mitigates that issue, by letting each user download and cache packages in a directory they own -- at the unfortunate potential expense of added disk use. Used together, these steps support group use of a shared conda installation; but, if you can spare the disk space, consider having users maintain their own e.g. Miniconda installations, to avoid coordination headaches.
  • It may sometimes be useful to create multiple development environments concurrently from the "same" code, perhaps based on different git branches. This would normally cause a conflict, as the same development environment name (e.g. DEV-hello) would be oversubscribed. You may set the DEV_ENV_PREFIX environment variable to change the DEV prefix to something else; for example, DEV_ENV_PREFIX=FOO make devshell would create development environment FOO-hello in the demo project.
  • Some of the techniques used in the Makefile, recipe/ directory, etc., of the top-level condev project are not meant for duplication in projects that would use the condev tools, but are required for bootstrapping in environments where condev tools are not yet available. The techniques used in the demo project are better examples to follow.
  • If recipe/conda_build_config.yaml defines multiple build variants, make devshell will choose the first variant returned by conda-build's api.render() function to define the packages to be installed in and available to the development shell. This may or may not be what you want, so you may need to temporarily simplify this file until the variant you want to interact with in your development shell is the one you get.