h
Thank you all for your reports and feedback. Currently having some trouble tracking this one down, but I can definitely reproduce it (as between 11.9005 and 11.9006). The odd thing is, as Carl reports, the performance in the later versions slowly gets worse, whereas it remains largely steady in 11.9005; I am not sure what might cause this sort of behaviour.
James, could you give me the tags, or the magic commit numbers, for the specific git commits which represent 11.9005 and 11.9006? That should give a finite list of commits to work through to figure out where it broke. (I can use the github "network" tool to go through them.)
If it's not obvious from the list of commits in between the two, then the next thing is to do a bisection: you try the commit halfway in between, and see if the bug is there.
You repeat until you find the commit which introduced the bug.
If there was a major merge of another branch in between those two, the commits just before and just after the merge are the first thing to check (because if the merge introduced it, then you have to go chasing back through the other branch).
Otherwise, it's pretty easy -- there's actually a command-line git tool designed precisely to help with bisection, "git bisect". If you don't know how to do a bisection, I might be able to do it for you (but no promises, not sure how much time I have). I'd need the test file, of course.