Manuel Drehwald

Can you share what file it's looking for, and how your Enzyme build folder looks like?

Ah good, only the file ending is wrong (so vs dylib), but at least the libenzyme version exists. Do you want to quickly test a fix and make a PR...

Windows is unfortunately unsupported anyway by Enzyme core, since MSVC dislikes Enzyme's interesting `#declare private public` ok, then I can make a quick PR.

Agree there are workarounds including also e.g. WSL2, but as long as the default Rust `x86_64-pc-windows-msvc` toolchain doesn't work due to this Enzyme core issue, I won't go around and...

windows-gnu should work (iirc. someone confirmed it in another issue here), but please feel free to test and confirm that if you have a machine at hand.

@jedbrown Using the new ENZYME_OPT dbg features as described [here](https://enzyme.mit.edu/index.fcgi/rust/Debugging.html#backend-crashes), I just got this mre: https://fwd.gymni.ch/7x2UOu

For case two, the union. I don't want to use the enzyme flag since I assume it would prevent Enzyme from catching some type errors. Could you introduce a TA...

> Probably I'd say to not mark this function with type tree's for the int if that's the case Thanks for the info, but that's the issue I have with...

> Maybe it would be helpful to add another debug tool here. We have the debug info so we could at minimum say "hey this is the source line the...

> > ssue I have with TA for Enzyme. I don't emit these TT myself, I only annotate the outermost function getting differentiated and a few mem* calls. And even...