Andy Polyakov

> I have also tried docker image based on "golang:1.18.8-bullseye", it is fine. So it sounds like a musl vs glibc thing... Not that it totally explains the failure on...

I'm not convinced that this is the same problem. The [original] problem is believed to have everything to do with how Go pointers are handled on the Cgo interface. Or...

Just in case for reference, what's up with internal batching. The algorithm in question requires temporary storage, by where to get is from? blst doesn't do heap allocations, so we...

> Somewhere within 1-2-3% performance penalty... Though I think that the evaluation was done with slower inversion subroutine... In other words it might turn out that reducing the internal batch...

> The part that is confusing to me is the function signature / double pointer indirection, why For flexibility. Note that it traverses the array of pointers till NULL, after...

> I don't think anyone has issue with internal batching The remark was more for the OP :-) As already said "just in case" :-)

> my ulimit -a shows 8MB Yes, but the main thread can start threads with smaller stacks... I'm not saying that it does, but it's the only thing I can...

Come on! It's not like we don't know it's missing :-) :-) :-) But on a serious note, there is a reason for why deserialization is not implemented. Because there...

See discussion in #93. "It might be helpful to view [blst] as something that works with EC points and byte sequences, rather than keys and signatures."

But the portable build comes with a run-time switch to ADX code path now[/already]...