numeric_big has been left out of parallel_schedule, requiring EXTRA_TESTS to run it, since going in back in 1999 (AFAICT it was even the reason EXTRA_TESTS was invented). The original commit states that it's huge, and it probably was. Today it runs faster than many tests we have in parallel_schedule, even on slow hardware like my ~5 year old laptop. Repeated runs in CI at various parallel groups place it at 50% the runtime of triggers.sql and 25% of brin.sql.
To make sure it's executed and not silently breaks, is it time to add this to the regular make check? -- Daniel Gustafsson