> I just ran the statprof unit test and it seems to still be OK. Perhaps > in Julien's run, there simply was not enough time in user space relative > to the sampling frequency. (Certainly statprof should deal with this in > a more intelligent fashion.)
That's certainly possible (I take it you're using "user space" to mean "non-primitive function code"), although I know it's collecting some samples at least because I've added `display' calls to statprof itself. One thing I just noticed is that the whole profiling process seems to work fine if count-calls? is false -- is that part of the unit test? On a more general note, it would be great if statprof were more robust / could profile more and different types of code. When you say "pull statprof into guile itself," Andy, do you mean getting profiling functionality into libguile or just including the module in the core distribution? (The former would be pretty sweet...) Regards, Julian
