jzmaddock

Still no reproduction here on: ``` uname -a Darwin dronerunnermac8 21.6.0 Darwin Kernel Version 21.6.0: Thu Sep 29 20:13:46 PDT 2022; root:xnu-8020.240.7~1/RELEASE_ARM64_T8101 arm64 ``` ``` sw_vers ProductName: macOS ProductVersion: 12.6.1...

Some questions: * What is the impact on compile time of using concepts rather than SFINAE or nothing at all? * Can we do "sophisticated" concepts that model for example...

Thanks Matt, just pulling down the code to experiment with now...

OK, just messing about with this, I'm seeing a small impact on compile times, which mostly seem to be coming from parsing the new header. I converted log1p.hpp as one...

MSVC build times for the same example are a bit variable, but it looks like around 3s with -DBOOST_MATH_DISABLE_CONCEPTS rising to 4s without.

Confirmed that error messages are much improved though: msvc is quite terse: ``` D:\data\boost\boost\t.cpp(12,26): error C2672: 'boost::math::log1p': no matching overloaded function found D:\data\boost\boost\t.cpp(12,33): error C7602: 'boost::math::log1p': the associated constraints are...

OK, I have a fix in the works for the second case, but the first is more problematic. The question is this: do we want to store and reset FPU...

>after https://github.com/boostorg/math/pull/883, with double promotion disabled, the call ibeta_inv(12.5, 1e-320, 0.9) no longer throws an overflow error, but the call ibeta_inv(1e-320, 12.5, 0.9) (a and b swapped) does throw. Addressed...

Can this issue be closed now, or are there still things to address?

>Yeah, I think getting it to work quickly with multiprecision is the tricky part. We would need the existing implementation for all non-builtin non-IEEE floats IMO. But it's not just...