Conda Error Troubleshooting¶

Errors plague the building of conda environments. The default behaviour of COBLE is to exit immediately on an error and give the opportunity to fix it. Additionally COBLE has built in by default build tools that are commonly the cause of compiler errors.

Generally it is best to install with conda if possible, and then install.packages/BiocManager/pip if not. However note that when moving up to a new version of R the packages may not have been created for conda so you then need to fall back to the native R. This may mean that to keep a recipe consistent fort some of the installs you may want to make it native R instead of conda. Or, change the blocks as needed, which is why r-conda and bioc-conda have been given as directives tio avoid name changes when maving package managers.

Here follows some common errors, why and how to fix them.

Initial errors¶

Sometimes initial errors indicate you are trying to create a conda environment on top of an old one. Do some aggressive cleaning, e.g.

# First, make sure you're NOT in the environment
conda deactivate
# Delete the environment
conda env remove -n my-env
# Or use the full path:
conda env remove -p /path/to/my-env
# And/Or use the file path
rm -rf /path/to/my-env
# Verify it's gone
conda env list

New R version - bioconductor not built¶

You may need to change from bioc-conda to bioc-package when a new version of r comes out, and potentially solve some dependencies. Some of the packages will not pull in dependencies from others, so for example Biocmanager will need you to explicitly make some conda installs. R-Forge packages will not pull in cran.

Missing compiler tools¶

Most are installed if you set the compile-tools and system-tools flag to true. Different versions of operating systems and code may have dependencies on old or newer compilers, so instead of just true you can specify a version eg 13.3.0.

For more bespoke compiler issues, you can use an environment variable in the flags, or set through bash, or explicitly set a more permissive compiler warning like:

bash:
  - CFLAGS="-Wno-error=incompatible-pointer-types" python -m pip install --no-build-isolation git+https://github.com/rachelicr/pysamstats.git

Error message about package corruption¶

SafetyError: The package for r-base located at /home/ralcraft/miniconda/pkgs/r-base-4.4.2-hc737e89_2
appears to be corrupted. The path 'lib/R/doc/html/packages.html'

Delete the corrupted package it will be recreated:

rm -rf /home/ralcraft/miniconda/pkgs/r-base-4.4.2-hc737e89_2

API limit¶

The error you will seee is {"message":"API rate limit exceeded for ...}. Or: HTTP error 403. API rate limit exceeded for 194.81.85.147. (But here's the good news: Authenticated requests get a higher rate limit. Check out the documentation for more details.)

This message will haunt you if you don't fix it properly. It is possible to get around it by installing from url and setting dependencies to False and then manually installing all dependencies. Far better in the long run to get a GitHub PAT token and set it in your bashrc.

To generate a token, go to Github-Settings-Developer Settings-Personal access tokens: https://github.com/settings/tokens. Geerate a new token (classic) for general use, and the scope that you need to select is all of the first one only, "repo". Once it is generated, copy it, and put it at the bottom oif your .bashrc like this:

export GITHUB_PAT="ghp_your-bash-token"

When building images or conda environments, the token will be used automatically by coble to authenticate you and avoid the API limit.

GLIBC compatibility issues (HPC)¶

It would be a last resoirt for me to skip a dependency I didn;t need rather than resolve it. COBLE is punitive and will fail until the error is ressolved by you one way or another (though you can choose to use the --skip-errors flag).

Error: package or namespace load failed for 'arrow' in dyn.load(file, DLLpath = DLLpath, ...):
 unable to load shared object '/data/scratch/DCO/DIGOPS/SCIENCOM/ralcraft/DEV/CBL/env-cbl/lib/R/library/00LOCK-arrow/00new/arrow/libs/arrow.so':
  /data/scratch/DCO/DIGOPS/SCIENCOM/ralcraft/DEV/CBL/env-cbl/lib/R/library/00LOCK-arrow/00new/arrow/libs/arrow.so:
  undefined symbol: __libc_single_threaded
Error: loading failed
Execution halted
ERROR: loading failed
* removing '/data/scratch/DCO/DIGOPS/SCIENCOM/ralcraft/DEV/CBL/env-cbl/lib/R/library/arrow'

To skip an unneeded dependency load the package manually without it, or use upi may be able to use conda:

bash:
  - Rscipe -e 'install.packages("tidyverse", dependencies = c("Depends", "Imports"))'
r-conda:
  - arrow

Missing header files (archived versions)¶

Trying to install an archived version for backwards compatibility can be hard. e.g. maptools is retired along with its dependecies. Modern installations of sp do not have header files so to get the installation you need to install version 1.6.0

using C compiler: ‘x86_64-conda-linux-gnu-cc (conda-forge gcc 15.2.0-16) 15.2.0’
using C++ compiler: ‘x86_64-conda-linux-gnu-c++ (conda-forge gcc 15.2.0-16) 15.2.0’
In file included from init.c:3:
rgeos.h:59:10: fatal error: sp.h: No such file or directory
   59 | #include "sp.h"
      |          ^~~~~~
compilation terminated.
make: *** [/home/ralcraft/miniconda/envs/452/lib/R/etc/Makeconf:204: init.o] Error 1
ERROR: compilation failed for package ‘rgeos’

I had to keep trying versions of sp until I found one that had header files and installed with R4.5.2 2.1.2 at https://cran.r-project.org/src/contrib/Archive/sp/sp_2.1-3.tar.gz.

Conda quirks and know-how¶

update message it can be annoying to get warning messages about updates when you just updated and the version is 0.01 out of date. You can supporess these warnings by passing a flag to the .condarc which you can do manually or by:

conda config --set notify_outdated_conda false