Update of Visual Studio C++ 16.2.x to 16.3.2 breaks some FTransform and physics computations

Development Programming & Scripting C++
1

Hi everyone

After having updated my Visual Studio 2019 from 16.2.x to 16.3.2, specifically the C++ compiler from version 14.22 to 14.23 some of my FTransform operations started giving me wrong results and physics started producing NaN results. Anybody else has similar problems?

Managed to temporarily solve the problem by installing the VC toolset 14.22 and setting WindowsPlatform.CompilerVersion = “14.22.27905”; in my Project.Target.cs file.

Not sure what is going on here. Maybe changes in the compiler the way vectorized (AVX) operations are produced?

Happens with official UnrealEngine builds 4.22 and 4.23.

2

I had issues with the FMath throwing errors on it’s checks.

I am not moving to 16.3 or beyond.

Not going to update any more unless Epic has announced it in their release notes.

This is what they announced in the 4.23 post.

IDE Version the Build farm compiles against

  • Visual Studio - Visual Studio 2017 v15.9.11 toolchain (14.16.27023) and Windows 10 SDK (10.0.16299.0)

    • Minimum Supported versions

      • Visual Studio 2017 v15.6

    • Requires NET 4.6.2 Targeting Pack