Specification of kernel directory on command line?

[mikecroucher]

Kernel specification files can currently be one of two locations - System wide or user specific.

Would it be possible to add the ability to specify a kernel directory on the command line? Something like

ipython notebook --kernel_dir=/Users/me/my_kernels/

This could take precedence over the other 2 locations.

One example where this would be useful:
I'm currently trying to build a portable Jupyter install on windows that includes an R kernel. It's for teaching use on a heavily locked-down managed desktop.

The install is based on the PortablePython project and works reasonably well. I have 2 problems, however:

• I have to install kernels to the user-profile which defeats the portability somewhat.
• If I move the directory containing the portable install, I have to regenerate my kernels.

The enhancement proposed here would fix everything since I could specify the kernel_dir as a relative path in a launch shortcut.

[takluyver]

Is sys.prefix within the portable install directory? You should be able to put kernel specs in sys.prefix + '/share/jupyter/kernels/' and it will find them.

[mikecroucher]

Yes, that works nicely. Thank you!
Furthermore, If I put the portable R folder within the directory specified by sys.prefix, this kernel appears to works fine:

{"argv": ["R/R-Portable/App/R-Portable/bin/i386/R.exe","-e","IRkernel::main()", "--args","{connection_file}"], "display_name":"Portable R" }

I'm going to test the portability by moving between machines but things are looking good!
Thank's for the swift and useful reply!

[takluyver]

Great. I'll leave this open because I think specifying a relative path in argv works by happy coincidence rather than deliberate design, and we might want to formalise a way to do that.

[mikecroucher]

Thanks. For info, here are the steps I took to get everything working using Portable Python and Portable R

http://www.walkingrandomly.com/?p=5734

[takluyver]

Great, thanks for writing that up. It looks like portable builds of Python will be an official thing when Python 3.5 comes out, so we won't have to rely on a not-very-maintained third party build. I quite like the idea of making 'Jupyter on a stick' - a USB stick with Jupyter, IPython, R and whatever else makes sense.

[takluyver]

jupyter/jupyter_client#8 would make it possible to specify an executable with a relative path from sys.prefix, like ['{prefix}/bin/R', ...].