Oh, if you are defining the problem as "pg_dump doesn't cope after I make this function in two steps" rather than "the server should allow this to be done in one step", that seems more defensible.
One could expect that that'd let pg_dump also cope with cases involving mutual recursion between two or more such functions, which is something we'd certainly not try to make the server allow without help.
It seems I was a bit too concise with my bug report. Sorry about this.
What I left out is this:
I have stumbled about this by accident. I had just been refactoring a
bunch of old sql functions and thought they might go well with sql_body.
The server didn't complain and all tests were passed until we came to
"dump and restore". That would have been a nasty surprise in an emergency.
I don't think it is a good idea to allow the creation of such a function
one way but not the other. This leads to confusion and unexpected
outcomes. Either allow sql_body with recursion or don't. There is already
a check if the objects exist that a function want's to use. So I thought it
must be easy to also check if a function calls itself. Changing how pg_dump
copes with this would at least lessen the impact of this situation.
