regions - Possible bug in RegionUnion or related functions?

Bug introduced in 11.2.0

---

Functions for derived geometric regions (RegionUnion and related) have been significantly updated in MMA 11.2 and now this simple example doesn't work any more as expected. Could this be a bug or I am using the functions in the wrong way?

d[radius_Real] := Disk[{0, 0}, radius];
r[size_Real] := Rectangle[-{size, size}, {size, size}]


$Version
Region[
 RegionUnion[
  RegionDifference[r[4.], r[3.]],
  RegionDifference[d[2.], d[1.]]
  ],
 ImageSize -> 200,
 PlotRange -> All
 ]

In version 11.1.1 I get the expected region shape.

And this is result of version 11.2. Also ToElementMesh fails to create a mesh from this region. (However, it does work if I replace the arguments of r and d with exact numbers.)

Answer

I think this is a bug.

Notice that applying BoundaryDiscretizeRegion to your region shows an error:

BoundaryDiscretizeRegion::defbnds: Unable to compute bounds for the region. Using default bounds of {-1, 1} in all dimensions.

Trying RegionBounds on it returns no result.

This explains the strange output.

However, changing RegionDifference[d[2.], d[1.]] to Region@RegionDifference[d[2.], d[1.]] "fixes" the problem. Now everything works fine.

Region does not seem to change the structure of RegionDifference[d[2.], d[1.]].

My conclusion is that this might be a bug. I would not expect different results when Region is or isn't used, only perhaps different performance (as Region is able to do certain checks a single time, and then guarantee that whatever it contains is a correctly defined region).

I suggest you report this to Wolfram Support.