[vtk-developers] ready to branch for 8.0.0?

[Hahn, Steven E.]

I've encountered a few complications with recent versions of MSVC and clang with integral value types.  With std::isnan and std::isinf Visual Studio 2017 generates an error message. With std::isinf<int> clang generates code that is 3-4 times slower than one would naively expect for a type that doesn’t contain infinity. I worked around both issues with tag dispatch, but don’t think we want that code copied and pasted throughout the VTK codebase.

Steven
________________________________________
From: Sean McBride <sean at rogue-research.com>
Sent: Wednesday, March 22, 2017 5:01 PM
To: Hahn, Steven E.; David Gobbi
Cc: vtkdev
Subject: Re: [vtk-developers] ready to branch for 8.0.0?

On Tue, 21 Mar 2017 15:41:00 +0000, Hahn, Steven E. said:

>The former. I’d like to move these functions (and workarounds for
>certain data types and compilers) into a header where they can be
>reused. Would it make sense to either template the existing function in
>vtkMath or put the templated functions in a different header file?

Maybe our new minimum compiler requirements would finally allow us to just use std::isnan, std::isfinite, and std::isinf.

Cheers,

Sean