I added a yolo "delete" statement to destroy an object that was being created a lot, probably couldn't be used anymore, and would *eventually* be deleted, and the problem went away for long enough for an order of more magnitude to be done before the inevitable segfault.
I think this means that if I can just clean it up responsibly, all will be well, but that might be outside the range of how much C++ I can pick up in a weekend.
If I reduce the amount of work to do by 90%, the job runs to completion and while the program still grows, the memory it allocates is freed, so I think this isn't a leak per se, just a lot of poorly managed lifetimes.
Comments
The second time, I asked for a memory dump without thinking -- yolo-pasting values from elsewhere -- and GDB crashed rather than try to allocate 30TB.