TBB path issue on Windows about bridgestan

[WardBrian]

I believe your diagnosis is correct. I have two thoughts:

1. Can we think of a reasonable way to test this behavior (it didn’t show up in our CI since we set $BRIDGESTAN, and I think we have to do that since the code will drift from the released versions)
2. I think adding it at load time is the more correct thing to do, since a model will (presumably) be loaded more times than it is compiled. If you compile, close python, re open, and then load, that needs to work.

On:

but if, say, someone would want to load the library using a different interface, that interface would then have to do that...

I believe that the dll_load_path is specific to Python as an additional security feature. To my knowledge $PATH is used elsewhere, so I don’t believe this would be an issue?

[aseyboldt]

I believe that the dll_load_path is specific to Python as an additional security feature. To my knowledge $PATH is used elsewhere, so I don’t believe this would be an issue?

I'll try with the branch from the PR. If this is in fact an issue, I guess we could also expose the function? ie as a verify_library_path or so, that only does something on windows?.