impl(gax-internal): use universe_domain when creating gRPC/HTTP transport

[alvarowolfx]

Towards #3646

[codecov]

Codecov Report

❌ Patch coverage is 96.55172% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.73%. Comparing base (df1db4d) to head (70e57b0).
⚠️ Report is 21 commits behind head on main.

Files with missing lines	Patch %	Lines
src/gax-internal/src/host.rs	92.98%	4 Missing ⚠️
src/gax-internal/src/http.rs	97.72%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5302      +/-   ##
==========================================
- Coverage   97.74%   97.73%   -0.01%     
==========================================
  Files         216      216              
  Lines       47482    47601     +119     
==========================================
+ Hits        46412    46525     +113     
- Misses       1070     1076       +6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dbolduc]

Aside, non-blocking: We don't need to define this more than once per crate.

And we could even put it in google-cloud-test-utils to define it once for all tests

[alvarowolfx]

Split gax changes into #5419

[dbolduc]

is private.my-custom-ud.com a thing? Or restricted.my-custom-ud.com?

In which case we would want to use like <service>.my-custom-ud.com as the host, but this would use private.my-custom-ud.com.

Duckie found b/361611906 for me. Seems like yes?

[dbolduc]

Don't be afraid to scrap this whole origin_and_header function. I wrote it and it was not my finest work.

(My memory is that I was trying to make it as efficient as possible, but (1) that is not so important for client-construction vs. like per request, (2) it made the code inflexible.).

[dbolduc]

I think this fails to strip the suffix on the .with_endpoint("foo.googleapis.com").with_universe_domain("my-custom.ud") case and then it returns "foo.custom.ud", when it should return "foo.googleapis.com".

Assuming I know what the caller is supplying to this function.

[alvarowolfx]

yeah, that's the conflicting thing that I'm stuck on this feature. You commented that before on #5302 (comment).

I'm trying to make the code look good and pass the test cases for localhost and the scenario that you described.

#[test_case("localhost:5678", "test.googleapis.com"; "emulator")]
#[test_case("https://localhost:5678", "test.googleapis.com"; "emulator with scheme")]
#[test_case("https://test.googleapis.com", "test.googleapis.com"; "GDU endpoint with universe domain") ]
#[test_case("https://test.another-universe.com", "test.another-universe.com"; "endpoint priority") ]
fn header_universe_domain(input: &str, want: &str) -> anyhow::Result<()> {
    let got = header(
        Some(input),
        "https://test.googleapis.com",
        "my-custom-universe.com",
    )?;
    assert_eq!(got, want, "input={input:?}");
    Ok(())
}

[dbolduc]

FWIW, I think these tests are wrong. Not sure why I wrote them.

google-cloud-rust/src/gax-internal/src/host.rs

Lines 161 to 163 in 822e215

	#[test_case("https://test.my-universe-domain.com", "https://test.googleapis.com"; "universe domain")]
	#[test_case("localhost:5678", "https://test.googleapis.com"; "emulator")]
	#[test_case("http://localhost:5678", "https://test.googleapis.com"; "emulator with scheme")]

Gemini tells me we should use "localhost" as the host when talking to an emulator, not the default service endpoint.

[alvarowolfx]

yup, I'm also finding that some previous assertions for universe domains might be incorrect, like

google-cloud-rust/src/gax-internal/src/host.rs

Line 159 in ffc6f3e

#[test_case("https://test.my-universe-domain.com", "https://test.googleapis.com"; "universe domain")]

, it should prioritize the endpoint override, so host should be test.my-universe-domain.com

[dbolduc]

Take inspiration from this:

https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=e3525652e19f18e517c6f64547d6a18e

The idea is just to give up if we don't recognize the universe domain as the GDU or the supplied universe domain option.

It fails if you have credentials for universe A, but then you try to make a request to universe B where neither are the GDU. Is that even a use case? It seems to defeat the purpose.

[alvarowolfx]

Split host resolution header into #5464