I didn't check out earlier versions of this patch, but the latest one still changes pg_size_pretty() to emit PB suffix.
I don't think it is worth it to throw a number of changes together like that. We should focus on adding pg_size_bytes() first and make it compatible with both pg_size_pretty() and existing GUC units: that is support suffixes up to TB and make sure they have the meaning of powers of 2^10, not 10^3. Re-using the table present in guc.c would be a plus.
Next, we could think about adding handling of PB suffix on input and output, but I don't see a big problem if that is emitted as 1024TB or the user has to specify it as 1024TB in a GUC or argument to pg_size_bytes(): an minor inconvenience only.
Last version still support BP in pg_size_pretty. It isn't big change. PB isn't issue.
We have to do significant decision - should to support SI units in pg_size_bytes? We cannot to change it later. There is disagreement for SI units in pg_size_pretty, so SI units in pg_size_bytes can be inconsistent.
There is no way at this point to add support for SI units in a consistent and backwards-compatible manner: both GUC and pg_size_pretty() use SI suffixes (kB, MB, GB, TB) with the meaning of 2^(10*n) (KiB, MiB, GiB, TiB). But given that the size relates to memory or disk space, it is quite customary *not* to use SI units, so I don't see a point in adding those.
