> Folks who are okay with LLMs think the regression test suite is the spec and is the guarantee of stability. How else can it be? If you are depending on some behavior not covered in the regression tests, how do you know the next minor release won't break you?

I think this deserves a real response.

First, an analogy. I drive along a cliff with no guardrail. How do I not drive off the cliff? By knowing how to drive. Sometimes people mess up and drive off the cliff

In practice people are operating Postgres as a machine, more than an abstract spec. Minor releases exist, but the changes are made by people operating the machine and who have a fear of making the machine break.

There are also performance characteristics that are part of an informal spec. While you can definitely write regression tests on performance, theres loads of value in stability of internals because people who have problems look into the machine and discuss it.

The internals might change, but there’s a lot of friction. So… you can have a lot of informal knowledge about.

If everyone working on a software stack is bathing in this informal knowledge, then the decision making is based on that. Things like what is meant to be in a minor or major release is understood. And people make those judgement calls.

After all, even if you have regression tests if you’re making changes you’ll need to write new tests! How do you know your new tests are right? That the new behavior is right?

PG is the entire machine. Fortunately we have version numbers, migration strategies, etc. But “here’s a new binary that passes the test suite and… maybe changes everything not covered by tests maybe doesn’t”. Why bother suffering when there are totally reasonable gradual rewrite strategies?

And of course… every bug in the world… got through despite a test suite! What software out there doesn’t have bugs?

And if the behavioral difference in the rewrite does affect people… I guess that’s something right? “Oh this isn’t a bug because it wasn’t covered in regression tests” is not that tenable.

>> There are also performance characteristics that are part of an informal spec.

They are not informal AFAIK - https://github.com/PGPerfFarm. A lot of these heavy duty OS projects have explicit Perf testing/nighlites (see Lucene's at https://benchmarks.mikemccandless.com/)

If you are saying the PG regression suite doesn't cover these perf tests - that is fair. I consider perf to be part of the "regression framework" generally.

>> How do you know your new tests are right? That the new behavior is right?

This is a more generic question in the LLM world. Reliable verifiers is what drives LLM loops. If you don't have these, you have no idea if you are actually making progress or you just generated code that returns 42 for every question. You needs something to actually ground the LLM output against. For existing features, the reliable verification is the existing regression/perf suites. For new features, the regression/perf suites should expand to fit. There are ways to go at this that range from ad-hoc (line/branch coverage) to fully formalized (verification-aware languages like Dafny).