Loss of precision with scattergl

[davidlhw]

Codepen link: https://codepen.io/DavidLHW/pen/MWzYXWK

In the above link, with small enough increments (in this case, at 1e-14), the plot will suffer a loss of precision if it's plotted using the scattergl type plot. Notice that the values are all clustered on the far left of the chart:

If plot type is changed to scatter, the data would then be plotted correctly like:

[nodakai]

@alexcjohnson Could you please take a moment to look into this issue? Unlike #5346, I've found that the issue of precision loss is specific to the markers component when using lines+markers with scattergl.

I'm using https://github.com/plotly/Plotly.NET for plotting time-series data that requires nearly microsecond-level resolution. With Unix timestamps currently hovering around ~1.7e9, IEEE-754 doubles serve my needs well, unlike IEEE-754 singles.

For my particular use case, I've found an effective workaround: I'm rendering high-frequency data streams via WebGL lines and overlaying less frequent, yet business-critical, "events" using non-WebGL markers. However, I recognize this approach may not be universally applicable.

[justinjhendrick]

I'm seeing a similar issue, but my x axis values are supposed to be integers. It seems that they're being converted to single-precision floats. Is it possible to force plotly and WebGL to keep the x-axis as integers to avoid the loss of precision?

from plotly.io import write_html
from plotly import graph_objs as go

##### Reproduction instructions:
# Use Python 3.11.6
# Use plotly 5.17.0
# Run this script with no arguments. It creates two Plotly HTML files in the working directory
# Open both HTML files in a web browser
#
##### Expected Behavior
# Red point locations in "with_scattergl.html" should be the same as "with_scatter.html"
#
##### Actual Behavior
# Red point locations in "with_scattergl.html" are incorrect.
# They are vertically stacked on the left side of the plot.


def plot(x, y, scatter, file_name):
    fig = go.Figure(
        data=scatter(
            x=x,
            y=y,
            mode="lines+markers",
            marker=dict(
                size=10,
                color="red",
            ),
            line=dict(
                color="blue",
            ),
        ),
        layout=dict(
            title=file_name,
        ),
    )
    with open(file_name, mode="w") as f:
        write_html(fig=fig, file=f)


def main():
    x = [
        1651196752467162363,
        1651196752469163377,
        1651196752569163988,
        1651196752669163486,
        1651196752769162766,
    ]
    y = [0, 1, 0, 2, 0]
    plot(x, y, go.Scattergl, "with_scattergl.html")
    plot(x, y, go.Scatter, "with_scatter.html")


if __name__ == "__main__":
    main()

[justinjhendrick]

The issue I pointed out in my previous comment is actually M1 Mac specific. See #6820