When I added support for nested ModelSpec computation with a shared factor cache, it broke the storage of transform state for Python factors with different formatting or nested transform states. We now pass through the entire transform state through to all model specs, even if it is superfluous to that model matrix. We can revisit this as an optimisation in the future, since it doesn't hurt to keep extra information.
When I added support for nested
ModelSpeccomputation with a shared factor cache, it broke the storage of transform state for Python factors with different formatting or nested transform states. We now pass through the entire transform state through to all model specs, even if it is superfluous to that model matrix. We can revisit this as an optimisation in the future, since it doesn't hurt to keep extra information.Closes #165.