Tim Blechmann

> Performance should be drastically improved on 64 bits builds out of curiosity: do you have numbers? slot-type dispatching will be much simpler with the tag enum than extracting the...

GC is always very slow. i was always wondering if it makes sense to increase the number of size classes to reduce the size of the lists for the individual...

> > back then when i introduced it to sclang, it gave a bit of a a speedup (maybe 5-10%ish, i don't recall the numbers). @timblechmann > > Are you...

i'd very much love to see support for user-provided `kevent`, similar to `posix::stream_descriptor` or `windows::object_handle`. afaict it's the only way to implement "event" or "semaphore" style functionality like with `eventfd`...

compare #1051

yes, the non-intrusive check is the challenging part. though i wonder if you have any thoughts about the approach that i've outlined above: my approach is intrusive into the `convert`...

> If the checking can be achieved like my example, it is very good. So trying non-intrusive version without msgpack-c library modification is good. unfortunately i don't see any way...

let me try to flesh this out in code and we can discuss in a PR. out of curiosity: i wonder what's the reason of using c++03 as minimum requirement...

> I guess removing C++03 support is harder. i guess it would be entirely appropriate to kindly ask people who cannot adopt c++11 in 2023 to use an older released...

@dyfer, you could double-check the nova-simd testsuite. the sign mask generation in https://github.com/timblechmann/nova-simd/blob/master/vec/vec_neon.hpp#L64-L69 should probably go through a union, though, as it may cause strict aliasing issues