Rewrite interactive API without holoviews dependency

[philippjfr]

This came out of some discussions with @jbednar on Friday. With the advent of more React-like APIs in Python we are in a place where we don't have a good way to declare standalone reactive variables, such as in this Solara example:

Initially I was thinking that we should be able to declare unbound param.Parameter instances which we could pass around but then I noted that we already have a very powerful reactive data structure in our ecosystem, namely the hvplot interactive API, which does not only support reactivity but also allows operator chaining and much more.

One of the main obstacles with .interactive is that it depended on dim expressions in HoloViews as a consequence of which it was specialized to support only objects that a holoviews.Dataset can wrap. Since I always had planned to rewrite .interactive to be independent of HoloViews AND to allow it to actually chain and store intermediates.

Unlike the hvplot implementation this implementation does not store the full chain of operations in hv.dim objects, instead we build up a chain of interactive objects each of which references the previous one. This means that intermediate results can be cached.

Migration plan

The plan is to implement a generic version of interactive here but allow it to be extended in all the ways that are required for hvPlot to handle plotting and add special handling for DataFrame and xarray Dataset/DataArray types.

Examples

Counter app:

count = pn.expression(0)
pn.Column(
    pn.widgets.Button(name='Click me!', on_click=lambda e: count.set(count.eval()+1)),
    pn.expression('Click count is {}').format(count)
)

ToDo

• Ensure all types of reactive objects are supported
  • param.Parameter
  • panel.widgets.Widget
  • Bound functions and methods
  • Other interactive objects
• Decide on top-level API, i.e. is it pn.ref
• Add tests
• Add documentation

[codecov]

Codecov Report

Merging #5138 (4a8d620) into main (2ff66b7) will decrease coverage by 2.49%.
Report is 1 commits behind head on main.
The diff coverage is 74.44%.

@@            Coverage Diff             @@
##             main    #5138      +/-   ##
==========================================
- Coverage   83.60%   81.12%   -2.49%     
==========================================
  Files         276      277       +1     
  Lines       40251    40387     +136     
==========================================
- Hits        33653    32764     -889     
- Misses       6598     7623    +1025     

Flag	Coverage Δ
ui-tests	41.36% <36.29%> (-0.11%)	⬇️
unitexamples-tests	69.98% <74.44%> (-2.48%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files Changed	Coverage Δ
panel/config.py	69.24% <10.00%> (-1.09%)	⬇️
panel/io/notebook.py	36.46% <20.00%> (-7.98%)	⬇️
panel/pane/ipywidget.py	51.89% <42.85%> (-1.95%)	⬇️
panel/pane/holoviews.py	79.25% <60.00%> (-0.32%)	⬇️
panel/depends.py	70.00% <66.66%> (-5.68%)	⬇️
panel/param.py	85.66% <70.86%> (-2.82%)	⬇️
panel/util/__init__.py	87.18% <80.00%> (-0.54%)	⬇️
panel/util/checks.py	87.87% <80.00%> (-0.65%)	⬇️
panel/widgets/base.py	92.10% <85.71%> (-0.42%)	⬇️
panel/tests/test_expression.py	98.46% <98.46%> (ø)
... and 5 more

... and 17 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[jbednar]

Sounds great to me; it's been really tricky to explain why .interactive is in hvPlot rather than in Panel. Just to be clear, with this support pn.interactive(df) would be roughly equivalent to hvplot's df.interactive()?

[philippjfr]

I have now split out an interactive_base which effectively only uses APIs that will be available in param 2.0 and interactive extends the base class adding all the Panel based display machinery. This means, in principle, interactive_base could live in param.