Facet sort order not respected with more than 5000 points

[rben01]

When the data has so many rows that vegafusion is required for Altair to produce a plot — I believe the max before vegafusion is necessary is 5000 — a custom sort order for facet columns is ignored. MRE:

import altair as alt
import numpy as np
import polars as pl
import vegafusion
from IPython.display import display

print(
    {
        "altair": alt.__version__,
        "polars": pl.__version__,
        "vegafusion": vegafusion.__version__,
    }
)

N = 5001
df = pl.DataFrame(
    {
        "person": np.random.choice(["Alice", "Bob"], N),
        "x": np.random.rand(N),
        "y": np.random.rand(N),
    }
)


with alt.data_transformers.enable("vegafusion"):
    chart = (
        alt.Chart(df)
        .mark_point()
        .encode(alt.X("x"), alt.Y("y"))
        .facet(column=alt.Column("person", sort=["Bob", "Alice"]))
    )
    _ = display(chart)

Version info: {'altair': '5.5.0', 'polars': '1.35.1', 'vegafusion': '2.0.3'}

Chart (with Alice before Bob, which is incorrect):

When you decrease N to 5000 (but keep vegafusion enabled nonetheless), you get the expected behavior:

[rben01]

I asked Claude about the bug. Unfortunately not only do I not know whether what Claude said is reasonable (at first glance, it looks like it could be) but even if I did implement it, I'm also not familiar enough with Vega to craft test cases. But in case it's helpful, here is what it said:

Fix: Preserve Custom Sort Order in Facet Charts

Bug Summary

Location: vegafusion-core/src/planning/lift_facet_aggregations.rs:82-277

When VegaFusion's lift_facet_aggregations optimization lifts aggregations from inside facets to pre-aggregated datasets, it completely ignores the custom sort order specified in mark.sort. This causes facets to appear in arbitrary order (often alphabetical) instead of the user-specified custom order.

Root Cause

The visit_group_mark function:

1. ✅ Correctly lifts transforms (formula, filter, bin, timeunit, aggregate)
2. ❌ Never checks mark.sort to add a corresponding collect transform
3. Result: Aggregated data has sort index columns but isn't sorted by them

The Fix

After line 237 (after adding the aggregate transform), add code to:

1. Check if mark.sort exists
2. Convert MarkSort to CompareSpec (extracting field and order from extra)
3. Create a CollectTransformSpec with the CompareSpec
4. Append it to lifted_transforms

This ensures the lifted dataset is sorted according to the custom sort order.

Implementation Steps

1. Add the sort preservation logic in lift_facet_aggregations.rs after line 237
2. Write a test case that verifies custom sort orders are preserved
3. Run existing tests to ensure no regressions
4. Format and lint the code