Don't leave empty dict containers behind when nested properties are set to None

[ahmojo]

Link to issue

Related to #5615

Description of change

Setting a nested property to None after it had been set could leave an empty compound container in the emitted figure JSON. For example, clearing marker.colorbar.thicknessmode left marker: {"colorbar": {}}. The same residue could be created when plotly_restyle(..., None) targeted a never-set nested path, even though no restyle message was sent.

This is inconsistent with construction, where empty compound containers are already pruned. This PR makes removal of a missing path a no-op and prunes emptied scalar compound containers while preserving list positions.

In the static-HTML reproduction from #5615, removing that residue prevents it from being included in the initial payload and avoids the subsequent JavaScript restyle collapse. This does not change plotly.js itself: a live FigureWidget whose browser-side state already contains colorbar: {} can still collapse after a later restyle.

Demo

I isolated the Python serialization trigger using the issue's Playwright reproduction (1200x800 viewport, the same bundled plotly.js for every variant):

Variant	Initial payload	Plot width after JS restyle	Colorbar thickness
Pristine figure	no colorbar key	543px	30px
Residual container from main	{"colorbar": {}}	64px	16290px
Same static-HTML repro with this PR	no colorbar key	543px	30px

With this change, the static-HTML output matches the pristine control. The separate live-FigureWidget path remains a plotly.js/frontend-state issue and is outside the scope of this cleanup.

Testing strategy

Added two regression tests in tests/test_core/test_figure_messages/test_plotly_restyle.py:

• Removing a never-set nested path leaves the figure unchanged and sends no restyle message.
• Clearing a nested property removes the empty scalar compound ancestors from emitted JSON.

Targeted checks:

• uv run --with-editable . --with pytest python -m pytest tests/test_core/test_figure_messages/test_plotly_restyle.py - 10 passed
• uv run --with-editable . --with pytest python -m pytest tests/test_core/test_graph_objs/test_property_assignment.py - 28 passed
• uv run --with ruff ruff check plotly/basedatatypes.py tests/test_core/test_figure_messages/test_plotly_restyle.py - passed

Additional information (optional)

This changes serialized output for set-then-unset scalar compound properties across the library, bringing assignment and restyle cleanup in line with construction-time behavior. Compound-array/list entries remain in place as positional placeholders, and batch updates are pruned when the batch flushes.

Guidelines

• I have reviewed the pull request guidelines and the Code of Conduct and confirm that this PR follows them.
• I have added an entry to the changelog if needed (not required for documentation PRs).

[camdecoster]

Thanks for the PR! We'll take a look and follow up. We're a bit backed up at the moment, so it might take a while to get through it.

[emilykl]

Hi @ahmojo, thanks for the contribution!

This is an interesting change. Pruning empty dicts left behind after a value is removed by setting it to None in plotly_restyle() does seem to make sense at first glance, although I'm not sure what the practical effects are beyond possibly this particular bug.

The reproduction example in #5637 never actually calls plotly_restyle() at all, since it uses the JavaScript Plotly.restyle() function, so I don't quite see how this change would affect the result of that particular example. Were you able to reproduce the issue in pure Python using plotly_restyle()?

[ahmojo]

Hi @ahmojo, thanks for the contribution!

This is an interesting change. Pruning empty dicts left behind after a value is removed by setting it to None in plotly_restyle() does seem to make sense at first glance, although I'm not sure what the practical effects are beyond possibly this particular bug.

The reproduction example in #5637 never actually calls plotly_restyle() at all, since it uses the JavaScript Plotly.restyle() function, so I don't quite see how this change would affect the result of that particular example. Were you able to reproduce the issue in pure Python using plotly_restyle()?

Thanks for pointing this out @emilykl. I tested this with a live FigureWidget using only Python. A direct restyle on a pristine widget works normally, but the same set-and-clear history from the issue reproduces the collapse:

widget = go.FigureWidget(make_fig())
display(widget)

Then, after it renders:

widget.data[1].marker.colorbar.thicknessmode = "pixels"
widget.data[1].marker.colorbar.thicknessmode = None

widget.plotly_restyle(
    {"marker.colorbar.thicknessmode": "pixels"},
    trace_indexes=[1],
)

The clear leaves colorbar: {} in the browser-side trace. The final restyle changes the thickness to 16,290px and shrinks the plot width from 543px to 64px.
This test also showed that my PR fixes the original static-HTML repro, but not this live FigureWidget path. I’ll adjust the description so it doesn’t overstate the scope or claim to fully close #5615.