[BUG] `dcc.Graph` `figure` prop may include Plotly.js `_inputArray` for typed arrays

[AnnMarieW]

When a dcc.Graph figure data contains typed arrays (for example int8, float32, etc.), the figure can include a Plotly.js internal field _inputArray

In the example below, the _inputArray is included after user interaction but not when the figure is first rendered.

_inputArray duplicates bdata and also includes decoded data, increasing callback payload size and overhead for large datasets.

Possible fixs:

• strip _inputArray in dcc.Graph before updating the figure prop. But that might be considered a breaking change since some people are now using _inputArray in callbacks.
• Have an option to not encode np arrays when using dash. This will also fix the issue of customdata not being available is clickData, hoverData if they are np arrays

Other related issues:

• Need a way to covert the typed arrays so the data is usable in a dash callback: [BUG] numpy array not correctly read back from figure data #3307

---

When app first renders (no _inputArray):

---

After hover (with _inputArray):

---

Sample App:

from dash import Dash, html, dcc, Input, Output
import numpy as np
import plotly.express as px
import json

app = Dash()

data = np.random.normal(0, 1, 250)
fig = px.histogram(data, nbins=20)

app.layout = html.Div([
    dcc.Graph(id="graph", figure=fig),
    html.Pre(id="fig-data")
])


@app.callback(
    Output("fig-data", "children"),
    Input("graph", "figure"),
    Input("graph", "hoverData"),
)
def debug(figure, hoverData):
    return json.dumps(figure["data"][0]["x"], indent=4)


if __name__ == "__main__":
    app.run(debug=True)

[AnnMarieW]

Here's a proposal for a function to decode a typed array in a dash callback:

import base64
import numpy as np

SHORT_TO_NUMPY = {
    "i1": np.int8,
    "u1": np.uint8,
    "i2": np.int16,
    "u2": np.uint16,
    "i4": np.int32,
    "u4": np.uint32,
    "f4": np.float32,
    "f8": np.float64,
}


def decode_typed_arrays(obj):
    if isinstance(obj, dict):

        if (
            "dtype" in obj
            and "bdata" in obj
            and obj["dtype"] in SHORT_TO_NUMPY
        ):
            try:
                arr = np.frombuffer(
                    base64.b64decode(obj["bdata"]),
                    dtype=SHORT_TO_NUMPY[obj["dtype"]],
                )

                if "shape" in obj:
                    shape = tuple(
                        int(x.strip())
                        for x in str(obj["shape"]).split(",")
                        if x.strip()
                    )
                    arr = arr.reshape(shape)

                return arr.tolist()

            except Exception:
                return obj

        return {
            k: decode_typed_arrays(v)
            for k, v in obj.items()
            if k != "_inputArray"
        }

    if isinstance(obj, list):
        return [decode_typed_arrays(v) for v in obj]

    return obj