manual kernel switching

[willwhitney]

While defining mappings between grammar names and kernel languages is good for cleaning up nonstandard language names, there are situations where you want to change the kernel being used manually:

1. Running some one-off code
2. Selecting which of several kernels for a particular language to use (e.g. python2 vs. python3)

To support these use cases, we should add some way of switching kernels.

I think it makes the most sense to replace the existing grammar -> kernel mapping config with this. The interaction might be something like this:

1. Open Command Palette
2. "Hydrogen: Select kernel for grammar " (Python Django)
3. Choose from all of the kernels, listed by their display_name (Python 3.4)
4. This mapping persists until it is changed again. Close Atom and reopen it to a Python Django file and it should load the Python 3.4 kernel.

Does this sound like a reasonable interaction?

[TKCen]

I'd say #40 is somewhat related, too. It would bind kernels to files though.

[Ardelean-Calin]

So is there a way right now to switch the ipython kernel from "ipython" to "ipython3"? Because it tries to run the ipython kernel ( which is ipython2), instead of the ipython3 kernel.

[TKCen]

I guess the solution would be to have ipython3 only on your Path, as the current kernel mapping seems to not support a mapping like:

{"Python": "python3"}

but maybe I'm missing something there. @willwhitney should know more.

[Ardelean-Calin]

Well, creating a symbolic link like this

ln -s /usr/bin/ipython3 /usr/bin/ipython

also works.

[willwhitney]

There are actually a few issues here.

1. IPython doesn't currently build kernelspec files for itself, meaning that Hydrogen can't know what ipythons are installed on your machine.
2. ipython kernel doesn't accept a connection file in the same way that other kernels do
3. Hydrogen needs UI for switching
4. and 2. mean that Hydrogen launches Python kernels differently than others; it always calls ipython. At present, I think using symlinks to make ipython call the IPython you want to run is the best option, as ugly as it is. If ipython currently points to a Python 2 ipython install, you can of course rename that to ipython2.

@rgbkrk is there a better way for Hydrogen to support this given the current state of the IPython kernel?

[rgbkrk]

On current master, the IPython kernel is installed as a standalone kernel with a kernel.json. This will be better on the coming 4.0 release, which is much better focused on supporting alternative kernels.

However, there is a way to install the kernels when on a 3.x release:

python2 -m IPython kernelspec install-self # Viva la Python 2
python3 -m IPython kernelspec install-self # Viva la Python 3

I'm on an ✈️, but I'll come back to this thread later. I certainly agree with needing a UI for switching

[stevengj]

As I mentioned in #119, this is also common for Julia developers who have multiple Julia versions installed for testing purposes. These are all installed as separate kernelspecs (julia-0.3, julia-0.4, and julia-0.5 in my case) in my Jupyter directory. What I would like would be:

• A way to change the kernel manually at runtime in Atom.
• A way to set a preference for the default kernel to use with a given language. (e.g. if I want all Python files to default to the Python 3 kernel, or all Julia files to default to the Julia 0.4 kernel)

[maximsch2]

Is it possible to also add an option of selecting a remote kernel? If I have an IPython notebook running on the remote server, it would be nice to be able to execute code in the context of the one of the notebooks there.

[willwhitney]

This has been discussed quite a bit. It's definitely on my "want" list, but
it's not clear that there's currently a good way to do this. I do expect to
implement this eventually, but don't hold your breath.

On Fri, Oct 9, 2015 at 6:22 PM Maxim Grechkin notifications@github.com
wrote:

Is it possible to also add an option of selecting a remote kernel? If I
have an IPython notebook running on the remote server, it would be nice to
be able to execute code in the context of the one of the notebooks there.

—
Reply to this email directly or view it on GitHub
#39 (comment)
.

[el-uhu]

Relating to this issue:
Hydrogen stopped working for me when, I upgraded to julia 0.4.0 (on Ubuntu 15.10 via ppa).
After rebuilding IJulia, normal jupyter notebooks worked fine. However Hydrogen would not connect properly to the kernel. This led to a situation where hydrogen would look like it had executed code, but in fact it only showed the ✔️ sign and produced no output.

I realised that the upgrade of Julia had deprecated the previous julia 0.3 kernel. This still showed up in jupyter kernelspec list but was no longer functional. Just deleting the deprecated kernel and reloading atom did not help, however re-installing hydrogen finally fixed it.

Whereas I realise that it bad practise anyway to keep deprecated kernels, I think exposing kernel-connection settings would definitely help.

[lgeiger]

See #138