In accordance with NEP-29, we have switched to have numpy 1.16 as the minimum supported version on all platforms.

We have changed the OSX and Linux platforms to enforce strict channel priority in package builds. This change means that if a package is available in the conda-forge channels, the conda solver will not consider any versions of the package from other channels. Users can disable this by setting channel_priority: flexible in their conda-forge.yml.

The main change is that openblas will use pthreads for threading by default on Linux instead of the previous openmp default. The openmp builds can be recovered by installing libopenblas=*=*openmp*.

conda-forge is moving to a new system for generating Core Dependency Tree (CDT) packages. These changes include:

• CDT packages will no longer be built using feedstocks and this practice is officially deprecated.
• Any current CDT packages in feedstocks will be moved to the new conda-forge/cdt-builds repo and the feedstock will be archived. Members of core will be doing this slowly on an as-needed basis, so it may not happen right away.
• Requests for new CDTs should be submitted as PRs to the conda-forge/cdt-builds repo.

These changes are being made so that conda-forge can provide access to CentOS 7 / glibc 2.17 for linux-64 builds. They will also move more of the packages needed for conda-forge builds into the conda-forge channels making builds more reliable.

conda-forge is moving to clang 10 on macOS! Check the release notes for what is new, breaking, or deprecated.

With CFEP-18 we now have a policy on how to deal with static packages. The most important change here is that we will be removing static libraries from the main packages and moving them to -static suffixed packages. -static packages will not be built by default but only on request.

The cf-mark-broken repo has been renamed to admin-requests. It still serves the same purpose. However, we have expanded the capabilities of the repo to be able to mark packages as not broken.

We are changing the way we mark packages as broken to better match the defaults channel and to better enable reproducible environments that depended on broken packages. We will now be adding the broken label to packages but leaving them on the main channel. In order to make sure they do not appear in the repodata.json for the main channel, we will be patching the repo data to remove them using the removals feature. Users will notice the following changes

• The packages on anaconda.org will now have both the main and the broken labels.
• All requests to mark packages as broken must be sent to the cf-mark-broken repo.
• Members of core can no longer mark things as broken by hand since the repo data patching must be done as well.
• The package metadata for broken packages may differ slightly from when they were on the main channel.
• The only correct source of package metadata is now the repodata.json etc on anaconda.org. Any other sources may be missing critical changes.

Starting this week, we are changing the way we upload packages to anaconda.org. We will move from direct uploads to the conda-forge main channel to using a staging organization/channel combined with a copy request from the staging channel to the production channel. This new process will allow us to perform some validation on the outputs of feedstocks before they are released.

What will you see as a feedstock maintainer?

• Starting this week, the admin-migrations service will be making commits to all feedstocks to provision them with the necessary configuration, API keys, and tokens.
• Each feedstock will now be provisioned with a secret token. This token should not be shared or taken out of the CI services. It is used to identify the feedstock during the upload process.
• The admin-migrations service will be setting a new top-level key in the conda-forge.yml, conda_forge_output_validation: true. This key indicates to conda-smithy that it should include the output validation calls in the feedstock CI scripts.
• Currently open PRs will need to have this key added by hand and then rerendered.
• When PRs are running the CI scripts, they will do some initial validation of the feedstock outputs. If this validation fails, the CI job will fail. Please see the CI logs for the error message which is printed after conda-build runs.
• Once a PR is merged to master, the copy from the staging channel to the production channel will happen automatically.
• Should a copy request fail, you will get a notification via a comment on the commit to master.
• As part of this process, uploads from appveyor will no longer be allowed unless there is a significant barrier to using azure. We have recently upgraded the compiler infrastructure on azure to support this change in policy.

Despite our extensive testing, we do not expect this change to be completely smooth, so please bear with us. As always, if you have any questions, concerns, or trouble, you can find us on Gitter or bump us directly on Github!

We are formally deprecating vs2015 in two weeks on 2020-04-07 and will move to vs2017. This change will enable us to support the usage of msbuild on Azure for the win platform and will provide additional support for C++. Most packages built with vs2015 can be linked with vs2017 toolchain (but not vice-versa). An exception is static libraries compiled with whole program optimization (/GL flag) which may be incompatible with the vs2017 toolchain. These static libraries will need to be rebuilt using vs2017.