I did run into problems by upgrading bokeh 2.4.3 to 3.0.3 by running pip install bokeh -U.
After the latest version of bokeh is installed and I open a Jupyter Notebook with very basic imports like
from bokeh.plotting import figure, show, output_notebook
the output shows that the loading process was successful.
But when I create a new figure and want to display it, there is no output. In the browser console there is an error that the bokeh-js version does not match the bokeh version.
The strange thing is, that the correct version of bokeh-js does exist, but is not loaded. For some reason not the latest nor a version based on the python version is loaded.
I did not install bokeh-js by any commands and I don’t know how to remove the old files. I tried to delete it without a positiv result.
At the moment my workaround is to create a new, empty environment, and install the latest version. This is working but not perfect. In fact I can reproduce this by downgrading from 3.0.3 to 2.4.3. With 2.4.3 it is working. If I upgrade again, the problem appears again.
I am working on an jupyter server with several environments and some of them I can’t easily remove and build again.
@mosc9575 Does this happen in brand new notebooks, or in old notebooks that explicitly had all their output cells cleared? The only time I have seen anything like this is re-running old notebooks across versions without clearing their output cells. There’s not really anything we can do about the way Jupyter behaves here.
That said, if this is happening on brand new notebooks, I have no explanation, and have definitely never seen anything like that. All I can suggest is to try to collect more data points (e.g. different systems, different python envs, etc).
Edit: do you have the jupyter_bokeh Jupyter extension installed? What version? Is this classic notebook or Juptyter-Lab? What version?
I run into the same problems using two different kernels with two different bokeh versions and I had the same problems. It turns out that the problem was not in bokeh and not in jupyter. I had to reload the browser to have no old java-script in the cache and the problem was solved.