Re: [bloom] Peer Review Conclusion

From: Kostas Savvidis (kotika98_at_[hidden])
Date: 2025-05-29 16:42:56

> On 29 May 2025, at 19:25, Vinnie Falco via Boost <boost_at_[hidden]> wrote:
> 
> Thank you for your work in the role of review manager, and also I think
> this sets a new bar for what we might like to see in terms of review
> summaries.

Thanks indeed go to Arnaud for managing the review. I am somewhat puzzled that no technical points whatsoever became part of the revew managers report.
The main technical innovation of the design of this library was the idea to do it essentially without hashing. This idea is solid, given
that a library like bloom is going to be dealing with millions or at most few billions of items. 
It is sufficient to use DETERMINISTIC random numbers, ie a good enough RNG.
The library is choosing this RNG parametrically which is unworkable and specific details are such that the choice may turn out to be really bad 
according to well established theory. 
The suggestion in my review was to use a Knuth 64bit RNG, but other choices are possible.
I do not imagine that there are people here who have not at least heard of the disaster awaiting the naive use of bad RNGs.

On a separate note, maybe we, the authors of the Boost.Random library should have done a better job of providing
RNGs which people might use in such a project instead of rolling their own. I am the author of the MIXMAX RNG
which is in Boost.Random, it is industrial strength, but this project needed something slightly more light-weight.
Unfortunately the small light-weight generators in Boost.Random and incidentally also std::random are simply no good. 
I.e. that 64bit RNG of Knuth is sadly not there afaik. 

Thus, we cannot fault the author for not using an RNG from Boost.Random, but not addressing this issue in the report at all is puzzling.
I hope that Joaquin finds a good way forward nonetheless.

Best Regards to All,
Kostas

=========================================
Institute of Nuclear and Particle Physics
NCSR Demokritos
https://inspirehep.net/literature?q=a%20Konstantin.G.Savvidy.1
https://mixmax.hepforge.org <https://mixmax.hepforge.org/>