test/goroot: classify precise GC and nil-check cases

[cpunion]

Summary

• classify thirteen GOROOT cases that require precise GC liveness, typed stack maps, or exact finalizer timing as not applicable to LLGo's conservative BDWGC runtime
• classify nilptr2.go and nilptr.go as not applicable because they require cmd/compile-style comprehensive explicit nil checks for derived or large-offset addresses that may not fault
• replace 37 platform/version-specific xfail or flaky selectors with 15 mechanism-based entries in notapplicable.yaml

Scope

BDWGC conservatively scans native stack and register roots. LLGo therefore does not provide Go's precise PC-specific liveness maps, typed roots that exclude pointer-shaped integers, typed cross-frame stack-object tracing, or exact finalizer timing. The thirteen GC and finalizer cases exercise those implementation properties rather than portable language behavior LLGo currently plans to reproduce.

LLGo preserves the required nil panic for an unused dereference in #2256. The remaining nilptr2.go and nilptr.go cases require a broader cmd/compile-style nil-check insertion strategy for address-only, derived-address, and large-offset operations. LLGo currently relies on LLVM and target memory faults for those forms and does not plan to reproduce that compiler strategy.

Parser and diagnostic errorcheck cases are deliberately left in xfail.yaml; this PR does not classify them as not applicable.

The test directives are unchanged and all cases continue to run. This only separates behavior outside the current compatibility goal from the remaining actionable xfail count.

Validation

• go test ./test/goroot -run '^(TestRepositoryExpectationsAreSeparated|TestNotApplicableMatch|TestXFailMatch)$' -count=1
• git diff --check

[fennoai]

Review summary

This PR reclassifies 4 errorcheck cases (fixedbugs/issue4776.go, issue13266.go, issue20789.go, bug050.go) from xfails in test/goroot/xfail.yaml to not_applicable in test/goroot/notapplicable.yaml, with reasons rewritten to the shared "not applicable" format.

Verified against the harness invariants:

• All 4 selectors were fully removed from xfail.yaml and appear exactly once in notapplicable.yaml, so the "appears in both files" check (runner_unit_test.go:155-158) and the runtime double-match guard (runner_test.go:355) stay satisfied.
• All 4 reasons satisfy TestRepositoryExpectationsAreSeparated (start with not applicable:, contain LLGo and compatibility goal).
• Pure data change; no matching-logic or performance impact.

One inline finding on accuracy of the bug050.go reason (verified against upstream source). No security or performance issues found.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

LLGo baseline benchmarks

0efe2c7ae904 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs base	Build	vs base	Run	vs base
Linux	cprintf	18656 B	+0.0%	257.954 ms	-1.6% (better)	1.069 ms	-0.1% (better)
Linux	fmtprintf	1885672 B	+0.0%	2.429 s	-0.9% (better)	2.757 ms	-3.1% (better)
Linux	println	68480 B	+0.0%	256.931 ms	-0.9% (better)	1.342 ms	+4.3% (worse)
macOS	cprintf	84672 B	+0.0%	357.312 ms	-11.0% (better)	3.005 ms	-43.7% (better)
macOS	fmtprintf	1890016 B	+0.0%	2.438 s	+5.4% (worse)	10.372 ms	-1.9% (better)
macOS	println	121216 B	+0.0%	312.251 ms	+1.1% (worse)	3.659 ms	+5.1% (worse)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs base
Linux	BenchmarkLookupPCRandom	11.320 ns/op	+1.5% (worse)
Linux	BenchmarkMergeCompilerFlags	137.500 ns/op	-0.4% (better)
Linux	BenchmarkMergeLinkerFlags	93.070 ns/op	-0.3% (better)
Linux	BenchmarkChannelBuffered	56.750 ns/op	+0.1% (worse)
Linux	BenchmarkChannelHandoff	25117 ns/op	-7.3% (better)
Linux	BenchmarkDefer	42.340 ns/op	+1.5% (worse)
Linux	BenchmarkDirectCall	1.008 ns/op	-13.1% (better)
Linux	BenchmarkGlobalRead	1.246 ns/op	+8.5% (worse)
Linux	BenchmarkGlobalWrite	8.438 ns/op	+0.0% (worse)
Linux	BenchmarkGoroutine	35598 ns/op	-2.7% (better)
Linux	BenchmarkInterfaceCall	6.251 ns/op	+0.7% (worse)
Linux	BenchmarkRuntimeGetG	1.630 ns/op	+0.2% (worse)
macOS	BenchmarkLookupPCRandom	10.620 ns/op	-0.9% (better)
macOS	BenchmarkMergeCompilerFlags	96.790 ns/op	-15.6% (better)
macOS	BenchmarkMergeLinkerFlags	72.730 ns/op	+5.7% (worse)
macOS	BenchmarkChannelBuffered	24.340 ns/op	+5.3% (worse)
macOS	BenchmarkChannelHandoff	8604 ns/op	+38.9% (worse)
macOS	BenchmarkDefer	29.420 ns/op	+2.0% (worse)
macOS	BenchmarkDirectCall	1.043 ns/op	-2.9% (better)
macOS	BenchmarkGlobalRead	1.090 ns/op	-0.5% (better)
macOS	BenchmarkGlobalWrite	1.062 ns/op	-2.2% (better)
macOS	BenchmarkGoroutine	26732 ns/op	-37.3% (better)
macOS	BenchmarkInterfaceCall	4.624 ns/op	+3.0% (worse)
macOS	BenchmarkRuntimeGetG	2.027 ns/op	+6.4% (worse)

Compared with 3f2ec409a461 measured in the same runner job.