Regression: Debugger loses all type information while stepping

I am having a weird bug (or a couple of bugs) with any version newer than 0.9.25: While stepping through the code, the debugger will lose all information about types/locals/etc. at certain points in the code. In other instances values of watched structures changes periodically while stepping (while the underlying data is constant as confirmed by prinf debugging and running it in the Visual Studio debugger).

0.9.25 does not show this behaviour, but any newer version does (I tried 0.9.26, 0.9.27 and the current master 16c447bb).

The executable was compiled with MSVC and was not changed between runs. I tried deleting the rdi file to no avail. I am running this on Win 11 25H2.

In this video you can see me starting the program, then stepping into the rollback_confirmInput function, then stepping line by line until I pass the assert(world) line at which point the debugger bugs out. I then continue running the outer loop up to the breakpoint, step into the function again, which behaves as before. Finally I watch the recordedInputs array of structs (which at this point should not change) and step line-by-line through the code during which you can see the values in the array change.

https://github.com/user-attachments/assets/b7a65a0b-7732-4e2a-b54e-3997622217ca

Here is the same sequence in 0.9.25, which behaves as intended:

https://github.com/user-attachments/assets/64c56e4f-3d63-401e-9d45-bffd0428e0e2

Tell me, if you need me to do a git bisect, to narrow this down further or supply other information. Cheers