This week I've worked on _d_interface_cast. I had to deal with some situations when implicit casts where made, but they were not lowered correctly. I narrowed down the issue to some code in expressionsem.d that created a new CastExp but did not perform the expression semantic analysis on it (where the lowering happens). There are still some failing tests I need to figure out, but it should be mostly done.

Apart from that, I have also run some benchmarks on the already templated casting hooks. Not surprisingly, the results show mostly on-par performance with the old versions, but in some cases the templated versions perform worse. I described more about this in a separate forum post.