While I don't personally have a use for it, I can see that someone will, and besides, it's kind of cool that you can do that ;) Question: would the interface be better specified with a constexpr initializer-list constructor rather than templating on an array of data?  That way you would have complete control of memory layout as well?  Or am I misinterpreting things? Best, John. On 21/01/2019 13:39, Ivan Matek via Boost wrote:

Hi everybody, my first message, I hope I do not mess this up.

I was contemplating writing a Compile Time Hash Containers library and I wrote a test implementation for sets and benchmarked it against mostly ska:: sets since they are the best.

Document is here:

https://docs.google.com/document/d/1IsUiWl3K0h2pgttYgMt67CxgJ3zeXBKZEA7hvAQ5...

It is quite long, but it aims to provide a introduction to the problem for people unfamiliar with details of hashing, and also it goes into a lot of discussions, if you care just about the numbers you can skip the "boring parts".

My interpretation of benchmarks is that it performs very well in some cases, decently in others, but I am biased. :)

Code is not available online since I do not want to dump a test project online without extensive documentation, but if somebody has a problem that may be benefited by this test implementation feel free to let me know, I would be happy to get some real world feedback on prototype. Limitations is that test implementation only works for sets, so no maps and if your objects are not integers you will probably not get much speedup. Also if your set has more than 100ish items GCC will run out of memory and VC++ does not compile code at all, so clang is the only way for most use cases today.

One more disclaimer is that I do not want this library to be some beast like ranges that takes years to implement, so if after reading the document you consider implementation quite basic I won't be offended :) In other words I feel that library like this although relatively simple compared to regular Boost libraries can provide quite a lot of benefit for some usecases where people really really care about performance of hash containers where values are known during compile time.

I know you can not predict the future of an unimplemented library, but I would be happy with some general feedback, and if you think this library would be nice boost proposal I would be interested in some pointers wrt how to proceed.

If somebody wonders how user code looks at the moment:

#include "cxhash_set.h" // also could be a result of constexpr function, no need for initializer list static constexpr std::array ints{0,1,2,3,5,8,21,34,64,128,256,512,1024,1701,1729,2048,4096,12345,123456,1234567}; int main() { cxhash_set<ints> ints_set; // contains could(should be?) static, but for now it is "normal" member fn std::cout << std::boolalpha << ints_set.contains(21) << std::endl; std::cout << std::boolalpha << ints_set.contains(22) << std::endl; }

P.S. feel free to comment in the google doc, I actually find mailing lists quite limited when it comes to prolonged and or detailed discussions. Maybe we can ping Alexandrescu for an instance of dlang forums, so we get a better discussion tools, and he gets a way to convert us to D. :)

_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

--- This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus

On Wed, Jan 23, 2019 at 4:34 PM Ivan Matek via Boost wrote:

I was contemplating writing a Compile Time Hash...library

The very first thing that popped into my head when I saw this is "I would like a library that emits a perfect hash function at compile time from a set of inputs." Motivating use case: https://github.com/boostorg/beast/blob/develop/include/boost/beast/http/impl... Regards