Gabriel Scherer

(This was sitting "unread" in my mailbox, so let me take the opportunity for a gentle ping. cc @dra27 )

No one spoke against the feature, but no one implemented it either. Anyone should feel free to implement it and submit a PR. Any volunteers?

This looks like impressive work, but it's also not so clear what it does right now from the code. Some early comments: - When you decide to un-draft this PR,...

Here is what I understand from the results: - we are comparing OCaml 4.14.1 with OCaml 5.2.0, and hopefully nothing else changed - the overall performance is noticeably worse in...

@ppedrot thanks! What are the GC settings for the 5.2 version?

I understand that the {native,vm}_compute machineries have been pessimized a bit in OCaml 5, which may suffice to explain the slowdown here. My hope is that there are *other* sources...

Ah, so this may in fact be related to major-GC tuning, and not anything specific about the vm_compute implementation. Good news, because experimenting with GC parameters is easier than understanding...

I looked at `major_words` results (the total number of words allocated on the major heap) for both runs of PCUICConfluence.vo (the file that sees a 40% memory blowup). They are...

We already have code that calls `/bin/sh` in C, for example in others/unix/execvp.c, and in OCaml, for example in otherlibs/unix/unix_unix.ml. Can we use/adapt this instead of adding more code?

I am nervous about `unreachable`. If I understand correctly, before we used `CAMLassert(0)`, which means "in debug mode have a clean error, otherwise do nothing", and then we would return...