FAQ

(Q) How to address permission denied (publickey) in circle builds?

When you see this error in a circle build:

Permission denied (publickey).

fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
Exited with code 128

Open a new issue with a comment @conda-forge-admin, please update circle. Once our web services have updated the circle configuration, restart the build.

(Q) Why does conda-build ignore the py37 selector in meta.yaml?

TL;DR: replace py37 with py==37.

conda-build has changed the selector syntax. You are now encouraged to use py==<version>, instead of py<version>. While the legacy selectors py27 and py36 are still valid, selectors py37 and higher are no longer valid.

Please use the new syntax py==27, py==36 and py==37 to avoid confusion.

Related links

(Q) What do build numbers above 1000 signify? How do I treat them?

TL;DR: there is no need for build numbers larger than 1000 anymore.

When you update a feedstock that still uses build numbers > 1000, the following rules apply:

  • When you increase the version, reset the build number back to 0 (e.g. 1005 -> 0).

  • When the version stays the same and you need to upload a new package, increase the build number by 1 (e.g. 1005 -> 1006).

Backstory: Build numbers of 1000 and larger are a relic from the compiler migration, where a build number offset of 1000 signified that a package was migrated to the new compilers. Since the completion of the compiler migration, this offsetting is not needed anymore. However, we cannot simply subtract the offset without updating the version, due to higher build numbers being preferred by the solver. Therefore build numbers above 1000 will gradually vanish as packages get updated to newer versions.

(Q) How to fix CMake not finding MSBuild.exe on Azure Windows builds?

TL;DR: Use Ninja or NMake Makefiles JOM as the CMake generator (cmake -G"Ninja"), and add build requirements for ninja or jom.

Sadly in the Azure Windows images, MSBuild.exe is not correctly setup for CMake builds with the Visual Studio generators. To work around this, you can use a different CMake generator, e.g. cmake -GNinja or cmake -G"NMake Makefiles JOM". These two are preferred because they allow for concurrent builds in contrast to e.g. only using cmake -G"NMake Makefiles"

(Q) Why does my new version appear on Anaconda Cloud, but is not installable with conda?

For certain, high-traffic channels (main & conda-forge), Anaconda uses a CDN to decrease costs. The CDN is only reindexed every 20 minutes, however Anaconda.org uses the original channel that the CDN mirrors. Therefore, packages will show up on the Anaconda Cloud about 20 to 40 minutes before they are downloadable via conda. You can use conda search <pkg> to see if the package is installable, because this command reads from the CDN.

(Q) How can I make debugging solver failures on the CI faster?

An alternative to the conda solver is the mamba solver. The mamba solver has a faster solve speed and prints better error messages that make debugging simpler.

You can configure the conda-forge CI to run a debug build using the mamba solver with the following changes:

  • Set build_with_mambabuild: True in conda-forge.yaml file

  • Rerender with conda-smithy (to rerender manually use conda smithy rerender from the command line)

Note

Builds made with mambabuild won’t be uploaded to conda-forge. These builds are purely for debugging purposes.

You can also do this locally by using:

  • conda install boa -c conda-forge

  • conda mambabuild myrecipe

For more details visit this page.

(Q) I am seeing Importing conda-verify failed. error message during build. What do I do?

Importing conda-verify failed. Please be sure to test your packages. conda install conda-verify to make this message go away.

You are seeing this error message because by default, conda-build uses conda-verify to ensure that your recipe and package meet some minimum sanity checks. This message can be safely ignored as conda-forge doesn’t use conda-verify.

(Q) When the bot creates a pull request to a feedstock to update the version, should I approve the pull request and wait with merging until everybody else that is a code owner has approved the PR?

There is no need to approve the PR. Every maintainer can verify and merge the bot PR without waiting on the approval of the other maintainers.

(Q) How to fix “build-locally.py fails with exit code 139”?

With Linux Kernel 4.11 there were some changes in the vsyscall linking. Depending on your distribution this may cause the above error. You can fix that on Debian by editing /etc/default/grub and specifiy GRUB_CMDLINE_LINUX_DEFAULT="vsyscall=emulate" in this file. Afterwards, you need to run update-grub and reboot your system. On other Linux distributions the fix is similar but you need to edit a different configuration file to change the Linux kernel cmdline. This workaround is only needed for images based on CentOS 6 (cos6). You could also workaround this by forcing the CentOS 7 based images using DOCKER_IMAGE=quay.io/condaforge/linux-anvil-cos7-x86_64 ./build-locally.py.

The exit code 139 itself actually is the general exit code for a segmentation fault. This could also mean that you have run into a different issue but the above issue is the most likely one with our CentOS 6-based images.

(Q) Is it necessary for me to be an upstream maintainer of the package I submit to Conda-forge?

Everybody can submit a package to Conda-forge, irrespective of whether they maintain the upstream version or not. Additionally, it’s not required but considered good practice to inform the upstream of a new package and invite them to be maintainers as well.

(Q) How do I fix the libGL.so.1 import error?

Error:

ImportError: libGL.so.1: cannot open shared object file: No such file or directory

To fix the error, create a yum_requirements.txt file and add mesa-libGL.

(Q) How can I fix the The Qt platform plugin "xcb" could not be loaded error during testing?

When testing packages that have a dependency on pyqt, the following error might occur under linux:

qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "" even though it was found.
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Available platform plugins are: eglfs, minimal, minimalegl, offscreen, vnc, webgl, xcb.

This comes from the CI environment being headless and can be fixed by adding the QT_QPA_PLATFORM=offscreen environment variable. The variable can either be added directly to the test command or provided in the meta.yaml like so:

build:
  script_env:
    - QT_QPA_PLATFORM=offscreen