I've been testing fexpr~ and it seems it won't get the previous samples from the previous block.
For example, a simple mean filter like this:
[fexpr~ ($x + $x[-1]) * 0.5]
will not work for the first sample of the block, because it won't get the last sample from the previous block.
So, I know that in practice it doesn't compromise much, it's just one sample, but in an 16 point average filter this becomes more of a concern. One way or another, even for just one sample, I consider this problematic and a bug, as it should always be able to address previous samples.