The recipes here create conda packages for scientific and numerical software components associated with the omnia
project.
The packages built from these recipes are shared with the community on anaconda.org.
These packages also depend on the conda-forge
conda channel.
To install a package (mdtraj
for example)
# Add conda-forge and omnia to your channel list, one time action
conda config --add channels omnia --add channels conda-forge
# Install the 'mdtraj' a package
conda install mdtraj
To install a package in one line:
# channels searched in order they are added here
conda install -c conda-forge -c omnia mdtraj
When setting up the configuration through conda config
, the channels are added to the top of the search priority
sequentially. So conda config --add channels omnia --add channels conda-forge
first adds omnia
to the top of the
list, then adds conda-forge
on top of that, giving conda-forge
the highest priority.
When temporarily searching through channels with conda install
, the channels are prioritized in the order they
are provided. So conda install -c conda-forge -c omnia
prioritizes conda-forge
first, then omnia
.
The Omnia project has started migrating to conda-forge
. New packages that
do not depend on OpenMM should be developed on conda-forge
and existing packages which do not depend on OpenMM
should start migrating if possible.
-
Update build image to CentOS 6 from CentOS 5
The base Docker image for linux builds will be updated to CentOS 6 with its new glibc. The base image is the
conda-forge
anvil, with some custom addons to include things like the AMD SDK, TexLive, and CUDA for GPU builds. The updated version will ensure packages can work on theconda-forge
platform which is CentOS 6 based. -
(Current) For packages that appear in
conda-forge
, remove the corresponding recipes inomnia
We want to minimize the amount of work we have to do as maintainers. To that end, we will stop building things which freely appear on
conda-forge
and maintained by someone other than us! For reproducibility purposes, we will keep our previously compiled versions, but they will not longer be updated.- Developers: if you want users to still exclusivley use the
omnia
conda channel, please see the Copying from conda-forge section below
- Developers: if you want users to still exclusivley use the
-
Allow recipes that do not depend on OpenMM to migrate from omnia to conda-forge
- Packages which do not depend on OpenMM and can be run on CPUs only should start migrating over to
conda-forge
in preparation for the total migration. - Packages which can compile with just the
conda-forge
linux-anvil should also start migrating. - We highly encourage devs of individual packages to start migrating now.
- Once a package is on conda-forge, it should no longer depend packages from omnia!
- Packages which do not depend on OpenMM and can be run on CPUs only should start migrating over to
-
Determine the appropriate way to build packages which require more than the
conda-forge
linux-anvil can provide- The
conda-forge
linux-anvil does not support some things such as some LaTeX packages, AMD SDK, and CUDA files. - We will need to reach out to the conda-forge people to see what the best course of action is
- The
-
Migrate OpenMM to
conda-forge
- This requires us to identify the best way to add CUDA and AMD APP SDK developer tools to the
conda-forge
linux-anvil docker image. We have some ideas, but no concrete solution yet.
- This requires us to identify the best way to add CUDA and AMD APP SDK developer tools to the
-
Move the remainder of packages to
conda-forge
- Also ensure that all former omnia packages can be installed without the
omnia
conda channel
- Also ensure that all former omnia packages can be installed without the
-
Change this repo into an archive for reproducibility.
PLACEHOLDER
It can be a bit confusing to rely on two conda channels where the order they are specified in changes which version of
packages are installed. During the migration to conda-forge
, developers can copy their binary tarballs from
conda-forge
to the omnia
channel using the Anaconda Cloud API, allowing users to rely only on the omnia
channel.
There are a couple conditions for this though:
- Packages still built in
omnia
will search for dependencies inconda-forge
thenomnia
, in that order - Your package should not depend on packages which only exist in
omnia
- You the developer will be responsible for also copying any dependencies from
conda-forge
toomnia
that you need and are not on the default channel
To copy packages:
If you have write access to the omnia
cloud
- Open an Issue
- This is important so we can track changes made to the cloud in a public space
- Note which package, version, and any dependencies you are bringing over
- Get the Anaconda CLI Tool
- Copy the package with the CLI
anaconda copy conda-forge/{PACKAGE}/{VERSION} --to-owner omnia
- Replace
{PACKAGE}
and{VERSION}
accordingly
- Copy any dependencies you need in the same way
- Close the issue
If you do NOT have cloud write access
- Open an issue, request which package and dependencies
- Request a maintainer who has cloud write access follow the steps above.
Eventually, all packages will be on conda-forge
and we won't have to worry about the multiple channels any more, until
then, we thank you for your patience as we go through this transition.
Python packages are built against latest two releases of python (3.5 and 3.6) and python 2.7. Packages which have a binary dependency on numpy are built against the latest two releases of numpy (1.10 and 1.11).
WARNING: Numpy 1.09 support will be phased out now that numpy 1.11 has been released.
The recipes here are automatically built using Travis-CI for linux
and osx
and Appveyor-CI for win
.
For linux
builds, we use a modified version of the
conda-forge linux-anvil,
to ensure that the packages are fully compatible across multiple linux distributions and versions.
This build image contains the additional tools:
- TeXLive 2016
- The CUDA Toolkit version 8.0
- The AMD APP SDK 3.0
There is an additional image which has clang 3.8.1
To build a package yourself, run conda build <package_name>
, or ./conda-build-all ./*
to build multiple packages across each of the supported python/numpy configurations.
- Fork this repo
- Add your
conda
recipe for building your packagepackagename
in a subdirectory calledpackagename
. Feel free to use other recipes here as examples. - Open a pull request to merge your branch into this master repo.
- It will automatically be tested to make sure it compiles.
- We will discuss the recipe and give suggestions about how to fix any issues.
- The recipe will be merged and our automated build framework will build
and deploy the packages to the
omnia
anaconda channel under therc
label. - Test the binaries by using
conda install -c omnia/label/rc packagename
- When you're sure the binaries are ready for a full release, comment on the
original pull request and a maintainer will move the package from the
rc
label to the main label.
Q: Should I include an md5
hash in my source:
section if using a Github compressed archive url:
?
A: No. Github compressed archives are frequently regenerated with different compression settings, etc., so md5
hashes cannot be trusted to be invariant. (#699)