Prevent segfault in distance to boundary calculation

[paulromano]

Description

This issue was discovered by a user. If a particle is traveling in a particular direction, leaves a cell with a complex region definition, and then goes to compute a distance to the nearest boundary and there are no boundaries in front of it (i.e., distance is infinite), right now this results in a segfault. This PR avoids that segfault with a one-line change. I've also added a quick test inspired by the user's original example that led to this.

Checklist

• I have performed a self-review of my own code
• I have run clang-format on any C++ source files (if applicable)
• I have followed the style guidelines for Python source files (if applicable)
• I have made corresponding changes to the documentation (if applicable)
• I have added tests that prove my fix is effective or that my feature works (if applicable)

[gridley]

Makes sense to me. So, if you have intersected on all surfaces and the distance is still infinity, this is taking the intersection surface to be any of the indices which led to infinite distance. Then it just causes the collision to limit the travel distance, right?

Now that this is allowed, maybe we should do a check around line 220 of particle.cpp that both the collision distance and intersection are not infinity. Otherwise it could cause an infinite loop (or it would go until it maxes out the particle event counter).

[paulromano]

So, if you have intersected on all surfaces and the distance is still infinity, this is taking the intersection surface to be any of the indices which led to infinite distance. Then it just causes the collision to limit the travel distance, right?

Almost -- the intersection surface is actually left unset (std::numeric_limits<int32_t>::max()). The collision will indeed be limited by the travel distance and all is good.

maybe we should do a check around line 220 of particle.cpp that both the collision distance and intersection are not infinity. Otherwise it could cause an infinite loop (or it would go until it maxes out the particle event counter).

Astute observation. In fact, if you take the test I put together here and change the air material to void, the code will still segfault because now there are no collisions. I'm honestly not sure what the behavior should be in this case. Basically, the particle is flying off to infinity and there are no boundaries to stop it. Open to thoughts on what we should do for that case.

[gridley]

IMO we should fatal_error at that point, maybe a message like "particle has left the Milky Way. exiting simulation."

[paulromano]

@gridley would you prefer that I handle the "particle leaving milky way" problem as part of this PR? I was going to follow-up with another PR handling that (along with a test), but I can also add it here if you'd like.

[gridley]

Sounds good to handle it in another, Paul!