I will try and get some test data with different vector sizes (made by Rust) tomorrow
On Wed, May 6, 2026 at 5:17 PM Micah Kornfield <[email protected]> wrote: > > > > I don't understand what you are proposing here? Are you saying that the > > spec would just hard code the number of elements? I think having it > > configurable is desirable > > > Yes, just hard-code. I agree it might be desirable but if we do make it > configurable, then we need to make sure the reference implementations > support this, and we have test data to verify cross-compatibility. > > On Wed, May 6, 2026 at 1:45 PM Andrew Lamb <[email protected]> wrote: > > > > There was some desire to have this configurable, but I wonder if > > we should postpone configurability for the sake of simplicity and > reducing > > the integration testing surface area. > > > > I don't understand what you are proposing here? Are you saying that the > > spec would just hard code the number of elements? I think having it > > configurable is desirable > > > > > Thank again Prateek for all your work here. > > > > > > 100% -- thank you Prateek > > > > On Wed, May 6, 2026 at 2:44 PM Micah Kornfield <[email protected]> > > wrote: > > > > > In reviewing the C++ implementation for ALP there are two things that > > stood > > > out in terms of testing that could impact the spec. > > > > > > 1. Currently vector size is hard-coded in the C++ implementation to > 1024 > > > elements. There was some desire to have this configurable, but I > wonder > > if > > > we should postpone configurability for the sake of simplicity and > > reducing > > > the integration testing surface area. > > > > > > 2. C++ is doing a fair bit of unsigned math. I haven't done a deep > dive > > > to understand the implications for how Java handles math, but we need > to > > > make sure Java is capable of handling extreme cases properly. > > > > > > Other than that I think it seems to be close to addressing > implementation > > > issues and spec wording. I'm looking forward to seeing this land. > > > > > > Thank again Prateek for all your work here. > > > > > > Cheers, > > > Micah > > > > > >
