Thanks Tom for clarifying promptly (and effectively helping me out), since that pointed me to find v9.6.6 related bug-fix around this, that is currently affecting an upgrade.
The v12 DROP does seem like an unrelated issue though, which I happened to notice coincidentally.
Robins Tharakan <tharakan@gmail.com> writes: > Is it expected for DROP TABLE CASCADE to find a related table but not drop > it? > The case in point, is when the base table is used as a column type. > This can at least be reproduced in v9.6 and v10, where it silently drops > the column!
Dropping the column is exactly what is supposed to happen: the scope of the dependency is defined to be just the column of that type, not the table containing it.
Not sure if this is documented anyplace in the SGML docs, but I'm pretty sure we have regression test cases for it.