Re: [lmi] [PATCH] Better support for non-x87 platforms

[Greg Chicares]

[See commit 571924bbc8f29c3002e28e4ee3f1c3f2c3269d22.]

On 2017-01-04 14:41, Vadim Zeitlin wrote:
[...]
> 1. The "Test floating point environment" menu item needs to be removed or
>    the handler for it reestablished or this test should be moved into the
>    GUI test completely.

Reestablished. For now, with x86_64 it just displays a message indicating
that it doesn't do anything yet. Later, we can write code to test that
execution path and the messages it produces.

> 2. fenv_is_valid() could be optimized if we want to avoid constructing
>    std::string/ostringstream unnecessarily.

I made a much smaller change. It seems sufficient just to show both the
rounding mode and the exception mask as ints. AFAICT, the expected value
for both is zero, so whatever value is unexpected will be obvious enough.

> 3. __BORLANDC__ checks need to be removed -- but it's not fully clear if
>    it should be done now or later.

Done.

> 4. Compilation of math_functors_test in 64 bits needs to be fixed, which
>    could be done either by making fenv_precision() always available (but
>    this would still require modifying math_functors_test.cpp, just in a
>    different way) or by just adding a preprocessor check.

I just conditionalized it, guessing that there wasn't much point in
comparing the accuracy of different precision settings if only one
setting is actually available.