BUG: out-of-bounds memory access when trying to zoom on a ParticleProjectionPlot

[neutrinoceros]

Bug report

Bug summary

Zooming may produce out-of-bounds errors. I'm not sure if this is specific to "SPH" datasets or to AREPO (or, none of the above).

Code for reproduction

import yt

ds = yt.load_sample("ArepoBullet")
p = yt.ParticleProjectionPlot(ds, "z",  [('all', 'particle_mass')])
p.zoom(10)
p.save()

Actual outcome

Traceback (most recent call last):
  File "/Users/robcleme/dev/yt-project/yt/ttt.py", line 6, in <module>
    p.save()
  File "/Users/robcleme/dev/yt-project/yt/yt/visualization/_commons.py", line 167, in newfunc
    self._recreate_frb()
  File "/Users/robcleme/dev/yt-project/yt/yt/visualization/plot_window.py", line 333, in _recreate_frb
    self._frb.render(key)
  File "/Users/robcleme/dev/yt-project/yt/yt/visualization/fixed_resolution.py", line 204, in render
    return self[item]
           ~~~~^^^^^^
  File "/Users/robcleme/dev/yt-project/yt/yt/visualization/fixed_resolution.py", line 776, in __getitem__
    add_points_to_greyscale_image(
  File "yt/utilities/lib/image_utilities.pyx", line 28, in yt.utilities.lib.image_utilities.add_points_to_greyscale_image
    buffer[i, j] += pv[pi]
IndexError: Out of bounds on buffer access (axis 0)

Expected outcome
no error

Version Information

• Operating System: macOS
• Python Version: 3.11.4
• yt version: 4.3.dev0
• Other Libraries (if applicable): numpy 1.25.0, matplotlib 3.7.1

yt installed from source

[matthewturk]

My guess is this is out-of-bounds-ing on the buffer access, not the pv access.

[matthewturk]

Probably worth clamping it and also seeing if the difference between the initial and clamped values is > 1. If we're looking at edge effects (which we might be since it's zooming in) then perhaps we just need a better system than calling <int>. It's possible that some particles may be included that are technically centered outside the bounds of the image.

[neutrinoceros]

I think that's exactly what's happening. The following patch resolves the issue

diff --git a/yt/utilities/lib/image_utilities.pyx b/yt/utilities/lib/image_utilities.pyx
index 7dc939ac4..5360ccf4e 100644
--- a/yt/utilities/lib/image_utilities.pyx
+++ b/yt/utilities/lib/image_utilities.pyx
@@ -25,6 +25,11 @@ def add_points_to_greyscale_image(
     for pi in range(npx):
         j = <int> (xs * px[pi])
         i = <int> (ys * py[pi])
+        if (i >= buffer.shape[0]) or (j >= buffer.shape[1]):
+            # some particles might intersect the image buffer
+            # but actually be centered out of bounds. Skip those.
+            continue
+
         buffer[i, j] += pv[pi]
         buffer_mask[i, j] = 1
     return

note that it's not expected to have much of a performance cost since the branching should be very well predicted (even in the reproduction script I've given, only one particle falls into the if clause)