Improve output of results_eq and results_neq when number of columns or their types are different

[wolfgangwalther]

It's a pain to debug right now, when you're only getting Number of columns or their types differ between the queries, but no additional info.

This patch just uses GET STACKED DIAGNOSTICS to show the original error message, which has the differing types included.

Not sure whether #37 would still be needed with this change.

[theory]

The challenge is that pgTAP currently supports Postgres 9.1 or higher and GET STACKED DIAGNOSTICS was added in 9.2, I believe.

[wolfgangwalther]

That's correct. That could be one of the reasons to drop support for 9.1, then... ;)

I can see that it makes sense to support older, discontinued versions of PG, as long as this doesn't block any improvements / development. But the last release for 9.1 has been in 2016...

In any case - if you still want to support 9.1, I can also put some conditionals in there to only enable this for 9.2+.

[theory]

Yes, let's keep 9.1 for at least one more release.

[wolfgangwalther]

Added conditionals for pg 9.1. I tried the following at first:

IF pg_version_num() >= 90200 THEN
   EXECUTE 'GET STACKED DIAGNOSTICS err_msg = MESSAGE_TEXT';
END IF;

But that results in a syntax error for GET STACKED.... So I wrapped the whole CREATE FUNCTION in a DO block and inserted the GET STACKED DIAGNOSTICS ... only for >= 9.2.

I was not able to spin up a local pg 9.1 to test this - let's see what CI says.

[wolfgangwalther]

So.. that works for pg 9.1. But, when I understand the output correctly, the update from 9.1 to 9.2 is broken. This must be because the function is not created again when upgrading. And since the function itself is not "dynamic" regarding the pg version, the pg9.1 function (without stacked diagnostics) is run on pg 9.2.

Any idea how to solve this?

[theory]

The way pgTAP has traditionally done this is to patch for older versions. So you'd leave GET STACKED DIAGNOSTICS in the source, but then add to compat/install-9.1.patch to remove it.

[wolfgangwalther]

So, I changed that and extended the patch. It does work the same way as before, but the update from pg 9.1 to 9.2 is still broken. And that makes sense, because running pg_upgrade does not add the GET STACKED DIAGNOSTICS back in. What am I missing to make this work?

[wolfgangwalther]

Ah, I can answer this myself now. Those cases are explicitly excluded in test/test_MVU.sh. I added the relevant file there, I guess it should pass now.