Support for Bokeh 3.4

[hoxbro]

I have chosen to hard-pin the build step. This can always be reversed.

• selectbox file was renamed to select in Allow Select to select values of any type bokeh/bokeh#13497.
• InputWidgetView._render_input() was added in Implement <label for="input"><input id="input"> bokeh/bokeh#13604
• URL dataclass was added in Improvements around BOKEH_DEV=true workflow bokeh/bokeh#13042
• ColorMap was renamed to PaletteSelect in Rename ColorMap to PaletteSelect and improve it bokeh/bokeh#13537

[hoxbro]

Else, it would give a type error.

[codecov]

Codecov Report

Attention: 8 lines in your changes are missing coverage. Please review.

Comparison is base (269b433) 84.19% compared to head (d0a03b1) 83.65%.
Report is 32 commits behind head on main.

Files	Patch %	Lines
panel/io/resources.py	70.58%	5 Missing ⚠️
panel/widgets/select.py	57.14%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6072      +/-   ##
==========================================
- Coverage   84.19%   83.65%   -0.55%     
==========================================
  Files         301      305       +4     
  Lines       45200    45518     +318     
==========================================
+ Hits        38056    38076      +20     
- Misses       7144     7442     +298     

Flag	Coverage Δ
ui-tests	39.64% <63.63%> (-1.09%)	⬇️
unitexamples-tests	71.65% <66.66%> (-0.18%)	⬇️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[hoxbro]

I can see the UI tests are starting to fail between Bokeh 3.4.0.dev4 and 3.4.0.dev5, this is likely to be because of bokeh/bokeh#13618, @mattpap do you agree? The PR is pretty big, and I can't see if it is backward compatible, with older versions of Bokeh.

A quick glance shows the following models are failing:

• Tabulator
• Plotly and Vega plots
• ReactiveHTML

[philippjfr]

Suspect this is due to the dict -> object/Map conversion. It used to be the case that most dicts would end up being converted to a JS Map but now you will end up with a standard JS object. Suspect those models will have to be updated.

[mattpap]

The PR is pretty big, and I can't see if it is backward compatible, with older versions of Bokeh.

It's not compatible on the implementation level. Previously if there was a dict in bokeh, then it was deserialized in bokehjs either as a plain object {}, if all keys were strings, or as Map in all other cases. Now dict is always deserialized as a Map. However, in all contexts, in particular all properties, that previously supported plain objects, now both plain objects and Map are supported. This is to keep API compatibility, while allowing preservation of object identity (no cloning) and performance (no proxying).

So what needs to happen, is that all cases where plain objects are used, instead of manipulate those objects directly, one should use APIs from core/util/object, like keys(), values(), entries(), dict(), to_object(), etc. Also don't ever use for .. in, but use for (const [key, vall] of entries(plain_object_or_map)).

This comes with some caveats. This approach breaks pseudo structs that are encoded as dictionaries, see e.g. bokeh/bokeh#13637. That issue is a critical regression to fix to make 3.4 viable. Otherwise I will need to revert PR bokeh/bokeh#13618. If you previously passed a plain object from a property to a third-party JS API, then now you have to use to_object() to make sure it's still a plain object.

[hoxbro]

Thank you, @mattpap, for the explanation.

Can you give an estimated time for when you look at the regression? (No pressure, just curious)

@philippjfr, I will wait with this PR until the regression or reversion has been made in Bokeh.

[mattpap]

Can you give an estimated time for when you look at the regression? (No pressure, just curious)

As soon as I'm done with UIs, so hopefully tomorrow.

[philippjfr]

Note I don't think the regression affects Panel, unless I'm misunderstanding.

[hoxbro]

Note I don't think the regression affects Panel, unless I'm misunderstanding.

I agree, but if the PR is reverted, there is no need to update Panel to support it.

[mattpap]

@hoxbro, I reverted dict/Map related changes, upgraded to dev8, fixed tsconfig.json (as mentioned in the meeting) and renamed one bokeh API call (rebuild -> update_children).

[mattpap]

I will publish removed commits in an new PR when this one is merged.