Do we need to require a user to have conda configured in rc files?

On some systems, it makes sense for users to rely on system python and only use conda for certain tasks. E.g. someone who spends a lot of time running old code may want a default environment that uses python 2, and rather than set up (base) in that manner is content to use the out-of-the-box python environment.

I removed the conda initialization from my .tcshrc file on the VM where I test using tcsh, and when I try to use jupyter-forward to connect I get and error about being unable to find jupyter:

$ jupyter-forward VMtcsh
********************************************** Authenticating **********************************************
Authenticating user (mlevy) from client (cgdm-chaohu) to remote host (localhost)
✅ The client is authenticated successfully
***************************************** Verifying shell location *****************************************
✅ Using shell: /bin/tcsh
************************************** Running jupyter sanity checks ***************************************
/bin/tcsh -c "which jupyter"
jupyter: Command not found.
*************************** ❌ Terminated the network 📡 connection to localhost ***************************

Is there some way to catch that the issue is actually that I commented out

# >>> conda initialize >>>
# !! Contents within this block are managed by 'conda init' !!
if ( -f "$CONDA_PREFIX/etc/profile.d/conda.csh" ) then
    source "$CONDA_PREFIX/etc/profile.d/conda.csh"
else
    setenv PATH "$CONDA_PREFIX/bin:$PATH"
endif
# <<< conda initialize <<<

(note that $CONDA_PREFIX is not defined without sourcing that file). One possible solution would be to pass an optional --conda-location argument; between that argument and self.shell maybe it would be possible to load conda before trying to do anything else?

ncar-xdev / jupyter-forward

Do we need to require a user to have conda configured in rc files? #143