On March 18, 2018 4:06:18 PM PDT, Tom Lane <t...@sss.pgh.pa.us> wrote: >Andres Freund <and...@anarazel.de> writes: >> I got a bit confused running installcheck-world and seeing ecpg >> failures like: >> ... >> A bit of pondering pointed me towards my environment's >> PGDATABASE=postgres being to blame. Unsetting that makes the test >> succeed. > >Hm ... pg_regress unsets PGDATABASE, along with the other related >environment variables, when it has a temp installation but not >when it doesn't. So what I don't understand is why your environment >doesn't also break every other regression test besides ecpg. >Perhaps they're all connecting to "postgres", but it's hard to >believe there wouldn't be conflicts if so.
All the others specify a database. The issue with the ecpg test is that it doesn't for two test cases. The failure is caused by libpq guessing the db name from the user name (hence the error about a database with role in it), which doesn't happen if libpq sees the environment variable. Andres -- Sent from my Android device with K-9 Mail. Please excuse my brevity.
