Thanks for the test case. The problem turns out to be sloppy coding in exprTypmod(): it recurses twice on the first argument of a COALESCE, making for exponential growth in the time needed to process a deep nest of COALESCE expressions ... which is exactly what your deeply nested FULL JOINs produce for the join USING variables. The patch attached fixes it for 8.0.* --- I've committed equivalent fixes as far back as 7.4, where the problem originated. Moral: sometimes an apparently trivial inefficiency isn't so trivial.
Interesting. Thanks Tom. You've saved me an unpleasant workaround.
Glad I could contribute *something*!
