Revert test serializer changes to avoid regressing existing use cases

[holly-cummins]

This reverts #40601, to fix quarkiverse/quarkiverse#94.
As part of the same changeset, I’ve also disabled a test that #40601 caused to pass (#40749), and enabled a test that #40601 makes fail (#40773).

Why is this a change we'd want, if it makes stuff fail? IMO, the stuff it makes pass is more important than the stuff it makes fail. The test which this PR now causes to fail has never passed, but the test which it gets to pass is reasonably widely used in the field. We don't want to regress it. How bad is the regression? Anything involving a Pact provider won’t work without this change. We should, IMO, favour ‘avoiding regressions’ over ‘enabling new functionality’. If 3.12 ships with a regression, that’s embarrassing, and if it has a regression that we actually knew about and decided we didn't care about, that’s something that will affect user trust in us.

Does this mean that #40601 isn’t a good change? No! It's an awesome change. This is a temporary reversion, just until junit-team/junit5#3820 makes it into a JUnit release. Think of it as a delay to function we definitely want, not a reversion. :)

This PR should be reverted as soon as we have a release of JUnit with junit-team/junit5#3820 in it. I’ve split the PR into two parts; we only want to revert 8e7f808, because hopefully the test enabled by other commit will continue passing. Once/if this merges, I'll open another PR with the reversion of the reversion (!), to make sure we don't forget.

There’s a fuller discussion of the arguments for and against this change on #40751 (which is an alternate solution to get these use cases passing).

cc @dmlloyd @geoand @gsmet @edeandrea

[dmlloyd]

🥲

[geoand]

😭

[edeandrea]

As much as I hate reverting stuff, I agree with the mindset and direction here. Better to not knowingly introduce a regression to something people use in favor of enabling something which has never been possible, especially knowing that there is a fix in the pipeline that will eventually make everyone happy.

[holly-cummins]

I got confused by my towering stack of disabled tests and test-related issues, and enabled the wrong test. Hopefully fixed now. :)

[gsmet]

I agree.

Let's monitor JUnit releases so that we can push it back as soon as we have a fix for the issue in JUnit.

[holly-cummins]

I've raised #40906 so we don't lose track of getting #40601 back in. We obviously can't merge it yet for all sorts of reasons, but it's there, with the right code changes for serialization.

@edeandrea, you can re-enable the superheroes tests now. 🥳

[edeandrea]

Done!

quarkusio/quarkus-super-heroes@d12cf5e

[geoand]

We now have JUnit 5.10.3 in main which includes David's fix, so I pressume we can bring #40601 back

[holly-cummins]

We now have JUnit 5.10.3 in main which includes David's fix, so I pressume we can bring #40601 back

(Almost) ready to go in #40906. I did it as soon as #40601 came out so we wouldn't lose track of it. :) Sadly, we have a failure in the 'fails with #40601' test that needs investigation.