Render matplotlib path collections (violin plots, pcolor, stems, etc.) in plotly

[robertoffmoura]

mpl_to_plotly silently drops several common plot types. Violin plots, pcolor, event plots, stack plots, fill_between and stem plots all produce PolyCollection/LineCollection artists in data coordinates, which the renderer skipped with "Dang! That path collection is out of this world...". The converted figure ends up with zero traces (except for stem, which gets partially drawn).

Fix: data-coordinate path collections are now drawn:

• collections with face colors (violin bodies, pcolor cells, stacked areas, fills) → filled polygons (fill="toself")
• collections without face colors (contour lines, stem lines) → plain line traces
• per-path colors/widths handled for collections that mix them; empty color arrays and degenerate paths skipped

Before: plots converts to 0 traces (empty figures).
After: plots render correctly.

Snippet to reproduce:

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np
import plotly.tools as tls

x = np.linspace(-3, 3, 30)
X, Y = np.meshgrid(x, x)
t = np.linspace(0, 2 * np.pi, 50)

cases = {
    "violinplot":   lambda: plt.violinplot(np.random.randn(100, 3)),
    "pcolor":       lambda: plt.pcolor(X, Y, np.sin(X) * np.cos(Y)),
    "eventplot":    lambda: plt.eventplot([np.random.randn(20) for _ in range(5)]),
    "stackplot":    lambda: plt.stackplot(np.arange(10), np.random.rand(10), np.random.rand(10), np.random.rand(10)),
    "fill_between": lambda: plt.fill_between(t, np.sin(t), np.cos(t)),
    "stem":         lambda: plt.stem(np.linspace(0, 2 * np.pi, 20), np.sin(np.linspace(0, 2 * np.pi, 20))),
}

for name, build in cases.items():
    fig, ax = plt.subplots()
    build()
    fig.savefig(f"{name}_mpl.png")

    p = tls.mpl_to_plotly(fig)   # 0 traces for all of these before the fix (except stem, which had some traces)
    p.write_image(f"{name}_plotly.png")

    print(f"{name:13s} -> {len(p.data)} traces")

Plot type	Converted plotly figure
violinplot (12 traces)
pcolor (900 traces)
eventplot (100 traces)
stackplot (3 traces)
fill_between (1 trace)
stem (22 traces)

[camdecoster]

Thanks for the PR! Overall this is in good shape but needs a couple of changes. Could you also please add a changelog entry?

[camdecoster]

Could you run the ruff formatter on this file?

[camdecoster]

This adds the opacity twice. convert_rgba_array already adds the alpha value. We don't handle this consistently right now, but I think this should be removed to avoid making the plot too transparent.

[camdecoster]

Should this be _export_color? If this should be export_color, you'll need to import it from plotly/matplotlylib/mplexporter/utils.py.

[camdecoster]

If i is greater than or equal to n, you'll get the same color for every call to per_path. To match the existing matplotlib behavior for the case of path length being greater than colors length, you could switch this logic a bit. This likely wouldn't happen much in practice, but this would cover us if it did come up.

Suggested change

	return colors[min(i, n - 1)] if n else default
	return colors[i % n] if n else default

[camdecoster]

This line needs to handle the case where x values are dates. You can see how this is handled for the same situation on line 456. This is actually done in a few places. If you want to extract a helper function, this could be used in all those locations.