New issue 1146: Octree construction on some SPH datasets revealing large swath of simulation space with zero density https://bitbucket.org/yt_analysis/yt/issues/1146/octree-construction-on-some...
Following the discussion in PR #1880, we identified that the octree construction was not behaving correctly for some SPH datasets. Rather than having full coverage in mesh fields across the simulation volume, regions of no gas particles have their mesh fields deposited as zero (i.e. zero density, zero temperature, etc.). Formally, this should never happen, and it can lead to a lot of unpredictable behavior later in processing the data.
For more information on this problem and how to reproduce it, see Pull Request #1880.