test/goroot: match malformed selector recovery

[cpunion]

Depends on #2222.

Summary

• match the exact primary diagnostic and exact g(f..3) source shape from syntax/ddd.go
• consume its three deterministic same-line follow-on diagnostics as independent one-use groups
• remove only the syntax/ddd.go xfail

The regression coverage was added before the allowance and reproduced the failure. It includes wrong-source, missing-primary, and wrong-line counterexamples; shared tests already cover exact message multiplicity and //line fail-open behavior.

Validation

• go test ./test/goroot -count=1
• syntax/ddd.go with an empty xfail set on Go 1.24.11, 1.25.0, 1.26.0, and 1.26.5
• all remaining 143 Go 1.26 compile/errorcheck xfails on Go 1.26.5
• git diff --check

No cross-line recovery allowance is introduced.

[fennoai]

Review summary

Reviewed the authored changes in test/goroot/runner_test.go, test/goroot/runner_unit_test.go, and test/goroot/xfail.yaml (the libslirp0 CI change and merge commit in the range are pre-existing, not part of this PR's work).

This is a well-scoped test-infra change. The parser-recovery pairing is deliberately narrow — it keys on exact primary message, exact source shape (parserRecoverySourceCode), file, and line, with one-shot consumption — so every GOROOT ERROR stays mandatory and unrelated diagnostics can't be silently suppressed. The diagnosticPathResolver.resolve change actually tightens the previous behavior: an absolute path now has to identify a configured source before it resolves, closing a small over-permissive fallback. The physical/hasLineDirective gate correctly falls back to leaving all recovery diagnostics visible when a //line//*line*/ directive makes physical-line pairing ambiguous. The six removed xfail.yaml entries line up with the deterministic secondaries now handled by the pairing table.

Verification performed:

• go vet ./test/goroot — clean.
• Ran the new/pure unit tests (TestParseDirective*, TestDiscoverCases*, TestRunProgramTimeout, TestRunGeneratedProgramUsesProvidedTimeout, TestValidateSystemMemoryState, TestShardCases, etc.) — all pass.

One minor, non-blocking observation is left inline. No correctness, security, or performance concerns found.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!