test: assert routines are created before the views and triggers that call them - #194

Why

The routine-ordering fix shipped in 3.0.0-rc.9 (#192) with no test of its own.

It was covered only incidentally — by a partition-trigger test that happens to need a function to exist. So a regression would have failed pointing at triggers rather than at ordering, and only on the Postgres legs. That is the kind of coverage that looks fine on a green run and misleads you on a red one.

I found this auditing my own work rather than from a failure, which is the better time to find it.

The bug it guards

CreateRoutine used to sort last, after CreateView and CreateTrigger, so a trigger whose function was also new was emitted before the function existed:

CREATE TRIGGER tg AFTER INSERT ON public.t ... EXECUTE FUNCTION f();
CREATE OR REPLACE FUNCTION public.f() ...

ERROR: function f() does not exist

Views have the same dependency, so both are asserted.

It earns its place

Verified the test actually fails without the fix rather than assuming it:

# DiffSorter reverted to the previous order
Failed asserting that 2 is less than 0.
FAILURES! Tests: 1, Assertions: 1, Failures: 1.

# fix restored
OK (21 tests, 35 assertions)

Also

The docblock on testUpOrderDropsProgrammableBeforeTables still described CreateRoutine as part of the trailing group with CreateView and CreateTrigger — which stopped being true when the fix landed. Stale comments that contradict behaviour are worse than none, so it now points at the new test.

Postgres suite 47 tests / 139 assertions; unit + SQLite 334 tests.

🤖 Generated with Claude Code