Chris K-L's recent complaint reminded me that we had been talking about
redefining int-to-bit(n) casting to take the rightmost n bits of the
integer value, instead of the leftmost n bits. The implementation
reason why it works the way it does is that the existing cast function
effectively converts to bit(32), and then after that we cast to bit(n),
and the bitstring width-conversion transformation takes the leftmost
bits of the bitstring, which is per SQL spec.
The only convenient way I can see to handle this is to extend the cast
stuff so that the cast function can be passed an additional parameter
which is the target typmod. Armed with that info, inttobit() could
align its output bits properly for the upcoming bitstring truncation.
Now that cast functions are selected through pg_cast, this should be a
fairly straightforward change. Does anyone have a problem with it?
I'm not sure the functionality is actually useful for anything except
this one issue, but arguably it's a general-purpose mechanism...
regards, tom lane