py_config unable to find vcruntime140.dll about reticulate

[jjallaire]

I'm not sure what's going on here. Some related discussion/threads:

ContinuumIO/anaconda-issues#443
http://stackoverflow.com/questions/33615829/vcruntime140-dll-error-after-creating-a-new-env-using-conda-for-anaconda-ide
http://stackoverflow.com/questions/40101662/python-win10-python35-dll-and-vcruntime140-dll-missing

[xarxziux]

This might be a problem with the way reticulate and Conda handle the PATH environment variable.

If I run source activate dl3.5 from Git Bash, the directory /C/Users/xarxziux/AppData/Local/Continuum/Anaconda3/envs/dl3.5 gets added to the PATH variable, so vcruntime140.dll and python35.dll become available within that Conda environment, but I'm guessing reticulate doesn't do this? Are you able to confirm if that might be the problem?

[jjallaire]

That definitely could be it! I just pushed a change to add the Python directory to the PATH for Anaconda on Windows: 2a552c6. Could you try again with this change?

[xarxziux]

I tried that, but I'm getting a completely different error message:

Error in python_config(python_version, required_module, python_versions) : 
  Error 1 occurred running C:\Users\xarxziux\AppData\Local\PYTHON~1\x64\2.6\python.exe 
In addition: Warning message:
running command '"C:\Users\xarxziux\AppData\Local\PYTHON~1\x64\2.6\python.exe" "C:/Users/xarxziux/AppData/Local/R-3.3.0/library/reticulate/config/config.py"' had status 1 

I'm not sure where it's getting the Python 2.6 directory from as I don't see any setting in the , but it's the first of several Python directories on my PATH variable, so it may have something to do with my setup. I tried re-ordering my PATH so that the Anaconda3 directory comes first, but the result was the same.

It looks like the script is reading the default Python location rather than the Anaconda3 location, although that may have more to do with my messy configuration. However, I see on the commit history for the last commit, you are checking for config$anaconda, but setting the dirname on the next line to config$python. Should that not be config$anaconda as well?

[jjallaire]

That script is just designed to to probing of various python versions (it actually runs against all of your detected configurations).

config$anaconda is just a boolean indicating whether it's an anaconda version of python. config$python is the path to the python binary.

I'm not sure what else to do given the complexity of your setup, let me know what are able to sort out and I'm happy to make any required changes to the package to make configuration succeed without special intervention.

[xarxziux]

Apologies for the complex set up :)

I don't actually use Python myself, but I've been tinkering with it recently so there's quite a few flavours on my machine. My goal here is to get RStudio linked with a Tensorflow-enabled Anaconda3 environment, which is what led me here. How do I tell reticulate where to find that environment? From what I can tell, it's pulling the data from the registry, in which case that's probably the official corporate location, but I've installed Tensorflow locally in a non-standard location. I've re-ordered my default PATH environment variable so that that location should be found first, but it's finding a 2.6 location that isn't even on my PATH. Can you please clarify how it's finding that?

Ideally I would like to be able to explicitly set the correct directory either by setting an environment variable, through a parameter in the use_condaenv() call, or by implying from the conda parameter, but I don't know how much extra work these would be. Other than editing the source code directly, do you know of a workaround for this?

[jjallaire]

You can set a RETICULATE_PYHTON (or equivalently TENSORFLOW_PYTHON) environment variable which will receive the highest priority in scanning. Next after that is the first Python on the PATH, next after that is any Python referred to by use_python, use_condaenv, etc., and next after that is scanning the registry for installed versions. It won't use a version that doesn't match the architecture of your R session (32bit vs. 64bit) and it won't use a version that doesn't include tensorflow.