Good ideas, and thank you for them! 
Further, I’ve realised now from the context around this issue, that it may well be a less-crashy version of this old issue.
And indeed, with some investigation along the lines that you suggested, I believe that I’ve found the source of the problem, and fixed it!
Specifically, it seems that there was an edge-case of sorts: When constructing a “room”, I was initialising its lighting-data–usually a list of values with a fixed length–to an empty list.
(As is fairly common for me when initialising lists–an empty list is often a valid “ground state”, I feel.)
Theoretically, this empty list would then be replaced by actual, proper data as the room was constructed.
But in some cases this just didn’t happen for one reason or another–most commonly at the moment, because the location in which the “room” exists has yet to be constructed, leaving it “blank”.
And of course, since the shader was expecting an array of known, fixed length, there was a clash when it was given an array of zero-length…
I’ve corrected this simply by switching the initialisation from an empty list to a list of the appropriate length, but containing zero (or near-zero) values. Which seems to work!