px.line erratic and wrong for large x value ranges

[eisenlohr]

Problem

Plotting a straight line from ($-X$, 0) to ($+X$, 1) for various (large) $X$ within a much smaller window results in wrong and inconsistent behavior, as demonstrated in below minimum non-working examples (using a Jupyter notebook with plotly.js v2.27.0).

I doubt that this behavior can be explained by intrinsic rounding problems, since

>>> np.interp([-0.1,0.1],[-1e17,+1e17],[0,1])
array([0.5, 0.5])

>>> np.interp([-0.1,0.1],[-1e18,+1e18],[0,1])
array([0.5, 0.5])

Examples

Data frame

df = pd.DataFrame(dict(x=[-1e17,+1e17,-1e18,+1e18],
                       y=[0,1,0,1],
                       xlimit=['±1e17','±1e17','±1e18','±1e18']))

Very tight plotting range (symmetric 0.1)

fig = px.line(df,x='x',y='y',color='xlimit',
              range_x=[-.1,.1],
              range_y=[0.4,0.6],
              width=400,
              height=400,)
fig.update_traces(opacity=0.5)
fig.show()

Slightly wider plotting range (symmetric 0.3)

fig = px.line(df,x='x',y='y',color='xlimit',
              range_x=[-.3,.3],
              range_y=[0.4,0.6],
              width=400,
              height=400,)
fig.update_traces(opacity=0.5)
fig.show()

Another plotting range (symmetric 1)

fig = px.line(df,x='x',y='y',color='xlimit',
              range_x=[-1,1],
              range_y=[0.4,0.6],
              width=400,
              height=400,)
fig.update_traces(opacity=0.5)
fig.show()

Another even wider plotting range (symmetric 3)

fig = px.line(df,x='x',y='y',color='xlimit',
              range_x=[-3,3],
              range_y=[0.4,0.6],
              width=400,
              height=400,)
fig.update_traces(opacity=0.5)
fig.show()

Apparently wide enough plotting range (symmetric 3.2)

fig = px.line(df,x='x',y='y',color='xlimit',
              range_x=[-3.2,3.2],
              range_y=[0.4,0.6],
              width=400,
              height=400,)
fig.update_traces(opacity=0.5)
fig.show()

As far as my tests go, for any range_x that is wider than about ±3, the plotting is rendered correctly:

fig = px.line(df,x='x',y='y',color='xlimit',
              range_x=[-1e17,1e17],
              range_y=[0,1],
              width=400,
              height=400,)
fig.update_traces(opacity=0.5)
fig.show()