relate_bnr.sql

[Paul Ramsey]

We’ve discovered that about half the tests in relate_bnr.sql are in fact wrong. That is, the results they immortalize are incorrect: the old version of RelateOp just game wrong answers for them. Because we’re testing GEOS behaviour in PostGIS, we now have a cross version issue. I believe that the most reasonable approach is to just remove relate_bnr.sql test file, as it’s mostly rotten.

If we leave the stable branches alone, that will mean that anyone upgrading GEOS underneath postgis 3.4 and lower will end up with regression failures. But again, that’s kind of hard to do anything about without major surgery. 

No questions to ask, really, there’s not a way to fix this. In general it again exposes the danger in locking in the behaviour of dependencies in our regression suite, which has bitten us in the past also, though never quite at this scale.

P.