[regression] Includes are no longer respected when POSTing/PATCHing

Hi killes! 👋 Cool to see you're using JSON:API :) Looking forward to hearing your unfiltered feedback about it: the good, the bad, the ugly.

That definitely sounds like a bug. Sorry about that! We'll need a regression test for this concrete scenario. There are lots of examples already in JsonApiRegressionTest. If you can write that test, I can promise a very fast fix. If you can't, no big deal, but then it might take a few days more.

I doubt it is specific to your setup. Rather than reproducing it manually, it's easier to write the test coverage in your place :) Reproduced!

Now that I think about this … I'm not 100% certain this is really a bug: it might be a feature request. https://jsonapi.org/format/#fetching-includes says this is something the server may support. Are you saying that this used to work on a previous version? If so, on which version?

This seems like it's something worth supporting. But it'd be helpful to know whether this used to work and we accidentally broke it, because then I'd dig in to the issue that broke it.

+++ b/tests/src/Functional/JsonApiRegressionTest.php
@@ -765,4 +765,46 @@ class JsonApiRegressionTest extends JsonApiFunctionalTestBase {
+   * Ensure POST and PATCH works for bundle-less relationship routes.

This comment still needs to be updated.

I did some archeology, and found that #2997600: Resolve included resources prior to normalization indeed broke this. If you run the test coverage in #8 against 4c23dce (the commit for #2997600) it fails, and in the preceding commit it passes.

Congrats, you're apparently the first person in over 3 months to use this feature of JSON:API 😀 👌

No time to dig in to the reason for that failure right now. It wasn't failing locally for sure.

I realized this also was a problem for PATCH, not just POST.

LOL. node entity type's weird permissions still trip me up after more than a decade. 😵

Cleaning up & make stricter.

This should fail.

Test-only patch in #17 is failing, #16 is passing. I don't see why I'd need to wait for @e0ipso or @gabesullice to RTBC/commit this since this involves a one-line fix + test coverage for a regression.

Moving to Drupal core's issue queue.

I'm working on https://www.drupal.org/project/drupal/issues/3122113