pg_restore V12 fails consistently against piped pg_dumps

Hi all,

When doing a parallel pg_restore (v12) against a dump created through a
pipe using an earlier version (11 all the way to 9.6), it fails with the
known error of not finding data offsets. I understand the reasons for it
(potential inability to seek), which is documented in pg_restore.

What I don't understand is why the same `pg_restore -j` worked in earlier
versions (say running pg_restore_v11 against the same dumps). Has anything
changed in terms of ordering? I am actually quite curious what led to this
finally breaking consistently.

I also timed things, and found that pg_restore_12 against a pg_dump_12
takes about the same time as pg_restore_11 vs pg_dump_11, on the same
dataset. I initially did this to see if for some reason pg_restore_v11 was
compensating for the lack of the offsets, but this didn't seem to be the
case (or perhaps the dump is too small to matter, around 1.5GB).

--
Wael