ssa: preserve package identity for unexported methods

[cpunion]

Fixes the shared method-identity root cause behind GOROOT fixedbugs/issue24693.go.

Summary

• qualify promoted unexported method symbols when their declaring package differs from the receiver package, preventing wrapper and ABI method collisions
• select interface method slots with the Go-compatible package-qualified method identity instead of comparing the spelling alone
• add the review counterexample as a real multi-package build-and-run regression test, update affected ABI goldens, and remove the Go 1.26 xfail

Tests

• go test ./ssa -count=1 -timeout=20m
• go test ./cl -run "^TestRunAndTestFromTestgo/(abimethod|interface|embedunexport-1598)$" -count=1
• go test ./internal/build -run "^TestBuildPreservesUnexportedMethodIdentity$" -count=1
• targeted SSA and multi-package regression tests with Go 1.24.11 and Go 1.25.0
• Go 1.26.5 GOROOT fixedbugs/issue24693.go with an empty xfail configuration
• new MethodSymbolName and iMethodOf branches are each covered at 100%

[fennoai]

Review summary

This is a well-targeted, correct fix. Qualifying an unexported method name with its declaring package path when the receiver package differs (MethodSymbolName) properly preserves Go's package-scoped identity for unexported methods, and the iMethodOf switch from bare-name to types.Id(pkg, name) comparison matches Go's interface-method identity semantics. Both symbol-producing sites (cl/import.go funcName and ssa/abitype.go abiUncommonMethods) route names through the same MethodSymbolName → FuncName pipeline and normalize via PathOf, so definition and reference symbols stay in sync. Test coverage is strong: unit tests for both MethodSymbolName and iMethodOf, plus an end-to-end fixture reproducing the collision, and the corresponding xfail.yaml entry (fixedbugs/issue24693.go) is correctly removed. The unchecked assertions (m.Obj().(*types.Func) on method-set selections, comma-ok elsewhere) are safe.

No blocking issues. Reviewers converged on minor/maintainability observations only.

Findings (non-blocking)

• ssa/interface.go:53 — removed // TODO(xsw): check signature. The removal looks intentional (interface method names are unique within an interface, so a signature check is unnecessary for correctness). Confirming this was an obsolete-cleanup rather than dropping a still-relevant concern.
• Performance (informational, not a concern): iMethodOf now computes types.Id(m.Pkg(), m.Name()) per interface method, which allocates one string per method for unexported names (types.Id is allocation-free for exported names). Bounded and correct-by-design for a compiler codegen path; no change needed.

See inline comments for the two maintainability notes worth a short code comment.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

LLGo baseline benchmarks

4d95cbf1d2b4 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs main	Build	vs main	Run	vs main
Linux	cprintf	18544 B	new	352.352 ms	new	1.366 ms	new
Linux	fmtprintf	2217608 B	new	3.296 s	new	2.746 ms	new
Linux	println	71504 B	new	355.001 ms	new	1.747 ms	new
macOS	cprintf	84672 B	new	300.827 ms	new	2.662 ms	new
macOS	fmtprintf	2361520 B	new	3.138 s	new	19.318 ms	new
macOS	println	125712 B	new	304.630 ms	new	3.120 ms	new

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs main
Linux	BenchmarkLookupPCRandom	12.260 ns/op	new
Linux	BenchmarkMergeCompilerFlags	144.400 ns/op	new
Linux	BenchmarkMergeLinkerFlags	93.910 ns/op	new
Linux	BenchmarkChannelBuffered	37.560 ns/op	new
Linux	BenchmarkChannelHandoff	23972 ns/op	new
Linux	BenchmarkDefer	43.740 ns/op	new
Linux	BenchmarkDirectCall	1.758 ns/op	new
Linux	BenchmarkGlobalRead	1.758 ns/op	new
Linux	BenchmarkGlobalWrite	2.811 ns/op	new
Linux	BenchmarkGoroutine	31683 ns/op	new
Linux	BenchmarkInterfaceCall	8.798 ns/op	new
Linux	BenchmarkRuntimeGetG	2.463 ns/op	new
macOS	BenchmarkLookupPCRandom	10.730 ns/op	new
macOS	BenchmarkMergeCompilerFlags	104.200 ns/op	new
macOS	BenchmarkMergeLinkerFlags	77.030 ns/op	new
macOS	BenchmarkChannelBuffered	24.310 ns/op	new
macOS	BenchmarkChannelHandoff	9474 ns/op	new
macOS	BenchmarkDefer	31.960 ns/op	new
macOS	BenchmarkDirectCall	1.136 ns/op	new
macOS	BenchmarkGlobalRead	1.029 ns/op	new
macOS	BenchmarkGlobalWrite	1.303 ns/op	new
macOS	BenchmarkGoroutine	28602 ns/op	new
macOS	BenchmarkInterfaceCall	4.702 ns/op	new
macOS	BenchmarkRuntimeGetG	2.612 ns/op	new

No main baseline exists yet; all metrics are marked new.

Warning

• The rendered benchmark data could not be pushed.