QuArK

Some more background on the "invalid floating point operation" errors: QuArK is written in Delphi, which by default enables floating point exceptions. This is not the norm, so a lot of code is never tested against this. This includes several OpenGL implementations, both on Windows and Linux. Affected OpenGL implementations can trigger a floating point exception during rendering, leading to the viewport reporting the "invalid floating point operation" error. This is not a Wine-specific issue: it has been known to occur on Windows and inside VMWare Workstation (Windows host + Windows guest).

Due to a copy-paste of assembly code from the Glide 2 programming guide, both the software and Glide (3Dfx) renderers in QuArK disable floating point exceptions when they are first initialized, and leave them off. This explains why using the software renderer before switching to the OpenGL/Direct3D renderer makes the error go away for that session.

The latest nightly builds of QuArK 6.6.0 Beta 7 contain a FP-exception workaround for the OpenGL renderer, enabled by default. Floating point exceptions are disabled when the call to OpenGL to render the viewport is made, and re-enabled when that call returns. But because this issue is restricted to OpenGL implementations (as this issue is not know to occur with Direct3D on Windows), the workaround is currently not implemented for the Direct3D renderer. In other words, if the used Direct3D implementation in Wine translates Direct3D-calls to OpenGL, this error can still happen. Note also that together with the implementation of this FP-exception workaround, the floating point exception disabling behavior of the software and Glide renderers has been corrected (i.e. removed), so that particular workaround no longer functions.

As can be seen from the test results, it appears that Wine started triggering this behavior between 3.0-rc4 and 3.0-rc5; my guess would be commit 5020fb7c478a080461df8655a3eac629648c4488. (This commit cannot cleanly be reverted on later versions of Wine, due to commit a83532d8dd5e99226dbccf0398083b052f033b9a, and probably more changed after that.) I haven't tested reverting this commit, as the commit is not the problem: the bug is in the OpenGL implementation, not QuArK or Wine. Also visible in the test results is that the error can sometimes disappear (my test results from Jan 22 2019 to Mar 08 2019 were good, but Mar 16 2019 is bugged again); this is probably due to the OpenGL implementation or drivers in the OS being updated or otherwise changed. Hopefully at some point the developers of these OpenGL implementations will correct their bugs, so that the floating point exception doesn't happen anymore in the first place.

(The title of this post should be:
The "invalid floating point operation" error explained
, but the website decided to eat it.)