this was what I was mentioning yesterday.
It's likely incorrect and incomplete :)
(I'm especially wondering about only computing the "interesting" regions' TC or also including regions outliving them... rn it's only the TC and passes our small test suite, and doesn't seem that much different from DatafrogOpt on rustc's tests. But last I checked, less than 20 failed but now it's 200)
In addition to incomplete, it might clash with the subset relations work, or future analyses, so I probably won't open a PR
For fun and motivation I ran the old NLL benchmark comparison, and the numbers seemed interesting in addition to probably wrong -- better to focus on completeness and correctness for the time being for sure
looks interesting, yes