[BUG] Off-axis projections and on-axis projections differ for oct-based datasets

[cphyc]

Bug report

Bug summary

Copied from #5059
OK - I've actually uncovered a funny one.

import numpy as np
import yt
import matplotlib.pyplot as plt
ds = yt.load("output_00080")
p1 = yt.ProjectionPlot(
    ds,
    "x",
    ("gas", "density"),
    center=[0.5] * 3,
    width=.8,
    #weight_field=("gas", "density"),
)
p2 = yt.OffAxisProjectionPlot(
    ds,
    [1, 0, 0],
    ("gas", "density"),
    center=[0.5] * 3,
    width=.8,
    #weight_field=("gas", "density"),
)

# Note: due to our implementation, the off-axis projection will have a
# slightly blurred cell edges so we can't do an exact comparison
v1, v2 = p1.frb["gas", "density"], p2.frb["gas", "density"]

fig = plt.figure()
plt.imshow(v1, origin="lower", norm=plt.matplotlib.colors.LogNorm())
plt.colorbar()
fig.savefig("/tmp/1.png")

fig = plt.figure()
plt.imshow(v2, origin="lower", norm=plt.matplotlib.colors.LogNorm())
plt.colorbar()
fig.savefig("/tmp/2.png")

fig = plt.figure()
plt.imshow(v1 / v2, origin="lower")
plt.colorbar()
fig.savefig("/tmp/3.png")
print(np.median(v1 / v2), np.mean(v1 / v2), np.std(v1 / v2))

This returns:

Projection	Off-axis projection	Difference

The funny thing is that the ratio between the on-axis and the off-axis projection is equal (within some uncertainties) to plot_width * 0.5675 (above, plot_width = 0.8). This is the case for a RAMSES dataset (here), but the same can be obtained using fake_random_ds instead, yielding the same value.

As much as I see how I could have missed a plot_width factor, what the heck is 0.5675?

Expected outcome

Both the projection and off-axis projection should be the same map (up to small differences).
This bug was most likely introduced in #4741.

Version Information

• yt version: 4.4 and above