You say the consequences of not fixing this bug are significant. What I'm saying is we cannot now break code in the wild that has already been written to accommodate those consequences. Sure, but that's true for virtually any bugfix so this isn't an absolute. This is a discussion with...
TypeError: 'NoneType' object cannot be interpreted as an integer or when I use thetf.random.Generator()class: ValueError: Attempt to convert a value (None) with an unsupported type (<class 'NoneType'>) to a Tensor. this is because I don't tell the model's Inputs layers the batch size...
@alextremblay, all packages must be a subpath of the project root, because otherwise it's not possible to put all packages in thesdist. And even if that would be possible, than the paths to that packages would be in thesdistdifferent to those in the system and one cannot build awheelf...