Uninferable method data sources share a discovery-failure test ID #7595

Summary

Multiple method data sources that cannot infer a generic test class's types produce discovery failures with the same test ID. Each failure should have its own identity so test runners can report every source independently.

Reproduction

On the generic-class method-source inference implementation at df2cf32:

public class Tests<T>
{
    public static IEnumerable<object[]> First() => [[1]];
    public static IEnumerable<object[]> Second() => [[2]];
    [Test, MethodDataSource(nameof(First)), MethodDataSource(nameof(Second))]
    public void Run(T value) { }
}

Use a targeted class filter with reflection discovery. BuildTestsFromMethodRows creates separate failure metadata for First and Second. DiscoveryFailureDataSource reaches the per-metadata failure fallback with a default TestDataCombination. GenerateFailedTestId includes class/method names and default combination indexes, but neither source identity nor the exception message, so both IDs are equal. This is established by the code path; an IDE streaming crash has not been reproduced locally.

Expected and actual

Expected: two separately identifiable discovery failures. Actual: both failure metadata entries derive the same test ID. Consolidate the per-method uninferred messages into one violation metadata entry with separately indexed data sources, or otherwise preserve source identity in the failure IDs. Retain constraint failures and sources whose declared inference returns null.

Environment

.NET SDK 10.0.403, net10.0. The same GenerateFailedTestId construction and per-metadata fallback exist on current main 69d3960; the new generic-class inference route is not on main yet. Triage against origin/main e6c7dee confirms that BuildTestsFromMethodRows still handles only ArgumentsAttribute; closed, unmerged #7571 is now replaced by #7596 and #7597. Related open/closed identity issues were searched; #7554 concerns an undiagnosed Windows session hang and is not evidence of this cause.

Reported as a nonblocking review finding while validating #7571: #7571 (comment). Please preserve a distinct failed-test identity for each source.

Acceptance criteria

• After Complete method-source generic class discovery parity and fix tuple normalization #7597 lands the new inference route, discover the two uninferable sources above with distinct, stable failure IDs (or one method-level violation with independently indexed sources).
• Preserve constraint failures and sources whose declared inference returns null; cover the default-combination fallback as well as source-indexed failures.
• Add focused regression coverage and verify both metadata modes for any metadata-collection change. Keep the uninferable-source policy from Complete method-source generic class discovery parity and fix tuple normalization #7597 unchanged.

Prerequisite: #7597, which introduces the affected discovery route. This follow-up is separate from parent #7224's discovery-parity acceptance criteria.