Reimar Döffinger

> It's 0xFF 0xFF, not two byes of 0 It's equivalent, initialize the crc to 0 and push 2 times 0xff or initialize to 0xfffff and push 2 times 0....

I am unable to see a connection between the test failures and this change...

I did not test on Windows. Only macOS Arm, Linux Arm, Linux x86_64 If you need to call the test function containing use of openssl function (vs. just importing/loading) then...

I think the issue and poor default comes from here, IMO promotion to long double is not appropriate on any architecture that is relevant today (anyone running on one where...

Hm, macOS for example does not even provide a long double type (at least on M1 and onwards - not sure for Intel), how do the tests work there? Changing...

Most tests use an epsilon relative to the type of the test, not the most precise type. But around 100 tests are broken as-is anyway when running on a platform...

Ah thanks! Now that I have a pass on the baseline, I will look into some of the failures with the promotion policy changed and see if I can help...

I sent https://github.com/boostorg/math/pull/1214 While I have not tested on x86, I think it should get you really close to change this define, at the very least on !x86

Seem it is kind of a duplicate of https://github.com/boostorg/math/issues/241 which is over 4 years old, I think it's really time to address it. It seems it is a very low...

Changing the default is easily possible now without breaking tests by applying the latest updated version of https://github.com/boostorg/math/pull/1214 Can then discuss how to best proceed with the change, e.g. leave...