Sam Westrick

Thanks, yeah, this is an issue we should fix. @qcfu-bu is it just the last comment in the file that is the problem? How is the formatting for comments otherwise?

> it requires the user to have a local sml installation. probably many users will anyway, but if they don't their experience will be limited. (or should we fall back...

I haven't explicitly tested for it yet! I assume it would probably be reproducible in MaPLe.

More progress: the quickhull correctness issue seems to have been a red herring. Previously, I was compiling MaPLe with `make smlnj-mlton`, but when I switched to standard `make` the correctness...

@darshand15, perhaps you could copy/paste the minimal reproducible example here? Just the actual source for the program I described above.

The problem isn't too severe because the scheduler will "fix itself" as soon as another task spawn occurs, at a par_do or parallel_for or otherwise. Any program with a large...

It would be really useful to try and actually trigger the bug! > Which operations are required to be safe for concurrent execution? Is it something like: each HM_chunk is...

This raises another question, about whether or not we currently have a performance problem due to the linking strategy.

(If you are interested in exploring this, a good place to start might be to add instrumentation to `HM_HH_getLevelHead` and `HM_HH_getLevelHeadPathCompress`, e.g., to count how many calls there are and...

> I was hoping that I could flag this with TSAN (and so mostly-mechanically surface all of the concurrent accesses that need to be synchronized), since IIUC, somewhat-distinctly from the...