Update field collection to allow inline fragment without type

[phughes-scwx]

Updates the algorithm for field collection to hew more closely to the version in the spec described in https://spec.graphql.org/June2018/#sec-Field-Collection. Specifically, it:

• only checks if the fragment type (for inline fragments) applies to the list of satisfying types if it is defined;
• handles the use of @include and @skip directives only after type application is performed (for inline and spread fragments).

This issue was identified in relation to the use of @defer on an inline fragment without a type condition, as described in #1972 . Because @defer is obscure and this was hard to recreate / test, this PR also adds an _example server that can handles deferred responses correctly.

I am not sure if there should be any documentation updates related to this.

I have:

• Added tests covering the bug / feature (see testing)
• Updated any relevant documentation (see docs)

[coveralls]

coverage: 59.754% (+0.2%) from 59.601%
when pulling 37acabd on phughes-scwx:issue-1972
into 1040b92 on 99designs:master.

[StevenACoffman]

Thanks! I'm extremely excited about the _example server addition, as it has been tricky to figure out how @defer is interacting with some other edge cases.

[StevenACoffman]

Odd that the Go 1.22 Integration / federation check failed, but the Go 1.23 did not. I'm assuming it's just flaky and not related to these changes, so that's annoying. (If anyone sees anything obvious, I would appreciate you telling me!)

[phughes-scwx]

Hm, yeah, odd. I'll continue to look. Are there any changes / upates you'd like me to make to the example or any docs? Or any additional tests for collection? If not, I'll remove the draft description after trying to pinpoint the 1.22 failures.

[phughes-scwx]

Looks like a similar issue to #2060 / #2409

[phughes-scwx]

@StevenACoffman after a few more passes and some tests, I think this is just flaky due to a slow startup, associated with the calls out to sleep in the integration test:

• https://github.com/99designs/gqlgen/blob/master/_examples/federation/start.sh#L23
• https://github.com/99designs/gqlgen/blob/master/.github/workflows/check-federation#L9
• https://github.com/99designs/gqlgen/blob/master/.github/workflows/check-federation#L19

We aren't really pausing for some meaningful signal, so might just get out of sync if the system is running slowly. (I could not replicate it.)

Are there any changes / updates you'd like me to make to the example or any docs? Or any additional tests for field collection? If not, I'll remove the draft description and hopefully the tests all pass when we trigger them again? 🤷

[StevenACoffman]

Thanks for looking at those flaky tests. It's annoying, but not this PR's concern.

Currently, for @defer the executable_schema.go has no test coverage for line 99 or lines 189-192 which means we cannot verify whether other changes accidentally break the correct behavior.

I'm not going to hold up merging this PR for that now, as it's wasn't your changes that caused those to be uncovered, but I would gladly merge any later PR that added more test coverage for them if you get a chance.

[phughes-scwx]

Yeah sure I'll do a follow up next week.