Re: « The PL/pgSQL interpreter parses the function's source text and produces an internal binary instruction tree... »

There does seem to be a general rule. But, as my example shows, there are exceptions to the rule. And it's impossible to make a simple user-facing statement of what determines "exceptional" status.

I suppose that the conclusion is clear: you can't be sure that a subprogram is good until every single code path (in the basic block coverage sense of this) has been tested. But, anyway, it was ever thus. (Error-free compilation never did guarantee error-free runtime outcomes.)

I'll call this "case closed" then.