Failures in Boost.Math misc test suite

[mborland]

The quaternion tests are now failing using cpp_dec_float (see here). Looks to be an inexact comparison between a floating point type and an integer.

Bernoulli tests also now fail using cpp_bin_float as it registers as an unsupported configuration (see here).

[ckormanyos]

quaternion tests are now failing using cpp_dec_float ... Looks to be an inexact comparison between a floating point...

Hi @mborland thenaks for finding this Matt!

Well, ... the fact that cpp_dec_float does not round has been present since the inception of cpp_dec_float.

Although this is not particularly nice, this behavior has not changed recently (although there is a completely separate, unmerged issue/branch related to this).

So I wonder what makes the difference in these tests now? Is it more exact knowledge or behavior of certain limbs when dealing with them within a constexpr context? Do we/I need to fundamentally change anything in cpp_dec_float? Or can we deal with that particular collection of test cases via tolerance increases?

Cc: @jzmaddock

[jzmaddock]

So I wonder what makes the difference in these tests now? Is it more exact knowledge or behavior of certain limbs when dealing with them within a constexpr context? Do we/I need to fundamentally change anything in cpp_dec_float? Or can we deal with that particular collection of test cases via tolerance increases?

I don't have time to debug this right now, but the failing tests really should have exact integer results IMO.

[jzmaddock]

Bernoulli tests also now fail using cpp_bin_float as it registers as an unsupported configuration (see here).

That's weird and rather perplexing, either I've made a mistake reconfiguring the Bernoulli code (but this did all pass I'm sure), or there's something up with . I might need to add an extra *info.cpp program to Boost.Config to probe the contents of so we can see what's up.

[ckormanyos]

what makes the difference in these tests now?

the failing tests really should have exact integer results IMO

Yes. I will try to reproduce the behavior and go forward with debug.

[jzmaddock]

Aaaahh, the Bernoulli failure isn't a failure - it's a test that is expected not to compile and the actual test does pass.

[ckormanyos]

Hi @mborland I believe @jzmaddock handled this by disabling a handful of cpp_dec_float tests in Boost.Math's miscsuite. See also math #712

Can we close?

[NAThompson]

Aaaahh, the Bernoulli failure isn't a failure - it's a test that is expected not to compile and the actual test does pass.

I got burned by this one as well. Is there a way just to not print anything on expected failures?

[jzmaddock]

Yes!

[jzmaddock]

I got burned by this one as well. Is there a way just to not print anything on expected failures?

Not that I'm aware of. The trick when searching for failures is to always search for "..failed" or "\.\.failed" if this is a regex (grep etc).

[NAThompson]

Ah I see. I always grepped for error.