R9: support independent Windows MSVC and MinGW profiles

[cpunion]

Part of xgo-dev/llgo#2325.

R8 (xgo-dev/llgo#2425) and native Windows calling-convention support (xgo-dev/llgo#2433) are merged. This branch is rebased onto the current xgo-dev/llgo main. All remaining LLGo modules use the released github.com/goplus/lib v0.5.0; no temporary fork replacement remains. The compiler fixtures retain the consolidated module layout from main.

Summary

• add independent native windows/amd64 MSVC and GNU/MinGW profiles without changing GOOS, GOARCH, Go flags, or the existing -target meaning;
• when CC is unset, probe the effective target of the default Clang: official standalone LLVM selects x86_64-pc-windows-msvc, while MSYS2 CLANG64 selects x86_64-w64-windows-gnu;
• keep optional CC, CXX, and -ldflags=-extld=... overrides Go-compatible, including quoted compiler commands and empty external-linker flag values;
• propagate the resolved physical triple and ABI through SSA code generation, C ABI lowering, libffi selection, linker/archive/import-library handling, dependency discovery, and package-cache fingerprints;
• reject non-Clang drivers, unsupported clang-cl syntax, architecture/ABI mismatches, and MSYS/Cygwin output targets rather than inferring the ABI from executable paths or the invoking shell;
• make the MSVC profile use official full-target LLVM 19, the Visual Studio SDK/UCRT, and native vcpkg dependencies without resolving compiler libraries from MSYS2;
• make the MinGW profile use CLANG64 LLVM 19, its GNU/UCRT C++ runtime, and isolated dependencies without inheriting Visual Studio SDK/toolset identity;
• provide profile-local pkg-config discovery for PowerShell/cmd and POSIX-style shells, so normal installed use requires no profile-selection environment variables;
• run both profiles through the existing Windows Go, LLGo, hello, demo, GOROOT, docs-local, build-cache, benchmark, runtime, and relevant coverage matrices;
• keep native exe, c-archive, and c-shared artifacts interoperable within each profile, including explicit MinGW COFF import-library generation;
• audit artifacts and environment identity so the MSVC lane cannot acquire msys-2.0.dll, cygwin1.dll, or libwinpthread, and the MinGW lane cannot silently reuse MSVC SDK/vcpkg state.

User-facing selection

No LLGo-specific profile variable is required:

GOOS=windows GOARCH=amd64 llgo build ./...

LLGo probes the default Clang target and appends the matching canonical target internally. Explicit CC, CXX, and external-linker settings remain optional Go-compatible overrides. CI verifies installed llgo.exe again with CC, CXX, PKG_CONFIG, and PKG_CONFIG_PATH unset from the normal profile shell.

Dependency boundary

The root compiler module and all demo/tool modules resolve the released github.com/goplus/lib v0.5.0. The old github.com/cpunion/lib replacements and the obsolete per-fixture modules are removed. The old standalone runtime/internal/clite/libuv/_demo module is also removed because it is neither runtime implementation nor test coverage.

Validation

After rebasing onto current main:

• range-diff preserves all 55 effective R9 commits and drops only four obsolete fork/fixture-isolation commits;
• go test ./cl -run ^TestRunAndTestFromTest(go|py|libc)$ passes on macOS ARM64 with Go 1.27;
• focused dependency-fingerprint and Darwin non-LTO/full-LTO/external-full-LTO integration tests pass;
• the full local run passes goroot, standard-library, C ABI, cross-compilation, SSA, and related packages;
• the remaining macOS TestMainGoexitLifecycleReleasedOnce rerun passed twice and hit the existing thread_get_state failed timing fault once; this file is unchanged by R9 and the fresh CI matrix is authoritative.

The pre-rebase R9 revision passed all 57 CI checks. The newly pushed revision reruns the complete MSVC/MinGW and non-Windows matrices.

Deliberate boundaries

• R9 qualifies both amd64 profiles. Native Windows ARM64 and WoW64 386 MSVC execution remain R10; their remaining GNU/MinGW qualification remains R11.
• MSYS2 and Cygwin are invocation shells, not target ABIs. Full Cygwin shell smoke and clean release packaging remain R12.
• clang-cl syntax and non-Clang drivers fail clearly until a complete driver-dialect implementation exists.
• Remote installation remains outside the Windows matrix; Windows docs coverage continues to use the local installation paths established by R8.

[github-actions]

LLGo baseline benchmarks

5c5f4a58b901 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs base	Text size	vs base	Build	vs base	Run	vs base
Linux	cprintf	19432 B	0 B / +0.0%	387 B	0 B / +0.0%	354.573 ms	+18.83 ms / +5.6% (worse)	1.421 ms	+71.86 us / +5.3% (worse)
Linux	cprintf-lto	19264 B	0 B / +0.0%	368 B	0 B / +0.0%	369.381 ms	+31.95 ms / +9.5% (worse)	1.480 ms	+117.4 us / +8.6% (worse)
Linux	fmtprintf	1652216 B	0 B / +0.0%	501874 B	0 B / +0.0%	2.733 s	+246.9 ms / +9.9% (worse)	4.028 ms	+315.1 us / +8.5% (worse)
Linux	fmtprintf-lto	1524168 B	0 B / +0.0%	460334 B	0 B / +0.0%	8.797 s	+226.8 ms / +2.6% (worse)	3.530 ms	-13.22 us / -0.4% (better)
Linux	println	62264 B	0 B / +0.0%	15286 B	0 B / +0.0%	354.217 ms	+6.601 ms / +1.9% (worse)	1.748 ms	+17.27 us / +1.0% (worse)
Linux	println-lto	54096 B	0 B / +0.0%	12914 B	0 B / +0.0%	539.191 ms	+24.42 ms / +4.7% (worse)	1.718 ms	+21.79 us / +1.3% (worse)
macOS	cprintf	84480 B	0 B / +0.0%	16669 B	0 B / +0.0%	501.429 ms	+56.24 ms / +12.6% (worse)	3.316 ms	+331.4 us / +11.1% (worse)
macOS	cprintf-lto	100704 B	0 B / +0.0%	16649 B	0 B / +0.0%	420.030 ms	-128.7 ms / -23.5% (better)	2.713 ms	-260.1 us / -8.7% (better)
macOS	fmtprintf	1512800 B	0 B / +0.0%	882816 B	0 B / +0.0%	2.703 s	+399.9 ms / +17.4% (worse)	12.844 ms	+7.747 ms / +152.0% (worse)
macOS	fmtprintf-lto	1208464 B	0 B / +0.0%	875856 B	0 B / +0.0%	7.969 s	+1.424 s / +21.7% (worse)	6.841 ms	+551.3 us / +8.8% (worse)
macOS	println	114784 B	0 B / +0.0%	35101 B	0 B / +0.0%	380.113 ms	-16.31 ms / -4.1% (better)	3.120 ms	-859.9 us / -21.6% (better)
macOS	println-lto	118656 B	0 B / +0.0%	32753 B	0 B / +0.0%	521.649 ms	-48.9 ms / -8.6% (better)	3.387 ms	-596.8 us / -15.0% (better)
Windows MinGW	cprintf	19456 B	new	4566 B	new	879.106 ms	new	3.530 ms	new
Windows MinGW	cprintf-lto	17920 B	new	4486 B	new	904.435 ms	new	3.495 ms	new
Windows MinGW	fmtprintf	1918464 B	new	591718 B	new	3.788 s	new	9.151 ms	new
Windows MinGW	fmtprintf-lto	1967616 B	new	583190 B	new	10.075 s	new	8.656 ms	new
Windows MinGW	println	74240 B	new	25062 B	new	871.376 ms	new	7.337 ms	new
Windows MinGW	println-lto	67584 B	new	21734 B	new	1.146 s	new	6.963 ms	new
Windows MSVC	cprintf	12288 B	new	4326 B	new	594.597 ms	new	3.563 ms	new
Windows MSVC	cprintf-lto	11776 B	new	4166 B	new	625.774 ms	new	3.494 ms	new
Windows MSVC	fmtprintf	1459712 B	new	591286 B	new	3.160 s	new	8.035 ms	new
Windows MSVC	fmtprintf-lto	1502720 B	new	589734 B	new	8.110 s	new	7.767 ms	new
Windows MSVC	println	47104 B	new	25062 B	new	610.013 ms	new	6.458 ms	new
Windows MSVC	println-lto	44032 B	new	22102 B	new	858.581 ms	new	6.091 ms	new

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs base
Linux	BenchmarkLookupPCRandom	12.300 ns/op	+0.05 ns/op / +0.4% (worse)
Linux	BenchmarkMergeCompilerFlags	144.400 ns/op	+0.4 ns/op / +0.3% (worse)
Linux	BenchmarkMergeLinkerFlags	94.850 ns/op	+0.61 ns/op / +0.6% (worse)
Linux	BenchmarkChannelBuffered	38.340 ns/op	-0.09 ns/op / -0.2% (better)
Linux	BenchmarkChannelHandoff	25101 ns/op	+1671 ns/op / +7.1% (worse)
Linux	BenchmarkDefer	47.270 ns/op	-0.91 ns/op / -1.9% (better)
Linux	BenchmarkDirectCall	1.758 ns/op	-0.351 ns/op / -16.6% (better)
Linux	BenchmarkGlobalRead	2.110 ns/op	+0.001 ns/op / +0.04742% (worse)
Linux	BenchmarkGlobalWrite	2.805 ns/op	0 ns/op / +0.0%
Linux	BenchmarkGoroutine	30530 ns/op	+477 ns/op / +1.6% (worse)
Linux	BenchmarkInterfaceCall	8.441 ns/op	+0.001 ns/op / +0.01185% (worse)
Linux	BenchmarkRuntimeGetG	1.759 ns/op	-0.001 ns/op / -0.1% (better)
macOS	BenchmarkLookupPCRandom	15.630 ns/op	+2.24 ns/op / +16.7% (worse)
macOS	BenchmarkMergeCompilerFlags	165.500 ns/op	+45.6 ns/op / +38.0% (worse)
macOS	BenchmarkMergeLinkerFlags	102.700 ns/op	+17.43 ns/op / +20.4% (worse)
macOS	BenchmarkChannelBuffered	28.620 ns/op	+1.28 ns/op / +4.7% (worse)
macOS	BenchmarkChannelHandoff	8762 ns/op	-844 ns/op / -8.8% (better)
macOS	BenchmarkDefer	42.710 ns/op	+6.94 ns/op / +19.4% (worse)
macOS	BenchmarkDirectCall	1.237 ns/op	+0.143 ns/op / +13.1% (worse)
macOS	BenchmarkGlobalRead	1.217 ns/op	-0.386 ns/op / -24.1% (better)
macOS	BenchmarkGlobalWrite	1.309 ns/op	+0.008 ns/op / +0.6% (worse)
macOS	BenchmarkGoroutine	38416 ns/op	+11819 ns/op / +44.4% (worse)
macOS	BenchmarkInterfaceCall	5.230 ns/op	+0.328 ns/op / +6.7% (worse)
macOS	BenchmarkRuntimeGetG	2.641 ns/op	-0.472 ns/op / -15.2% (better)
Windows MinGW	BenchmarkLookupPCRandom	14.460 ns/op	new
Windows MinGW	BenchmarkMergeCompilerFlags	557.100 ns/op	new
Windows MinGW	BenchmarkMergeLinkerFlags	465 ns/op	new
Windows MinGW	BenchmarkChannelBuffered	39.020 ns/op	new
Windows MinGW	BenchmarkChannelHandoff	1543 ns/op	new
Windows MinGW	BenchmarkDefer	54.130 ns/op	new
Windows MinGW	BenchmarkDirectCall	1.747 ns/op	new
Windows MinGW	BenchmarkGlobalRead	1.752 ns/op	new
Windows MinGW	BenchmarkGlobalWrite	2.792 ns/op	new
Windows MinGW	BenchmarkGoroutine	75986 ns/op	new
Windows MinGW	BenchmarkInterfaceCall	10.140 ns/op	new
Windows MinGW	BenchmarkRuntimeGetG	2.101 ns/op	new
Windows MSVC	BenchmarkLookupPCRandom	9.451 ns/op	new
Windows MSVC	BenchmarkMergeCompilerFlags	421.500 ns/op	new
Windows MSVC	BenchmarkMergeLinkerFlags	372.900 ns/op	new
Windows MSVC	BenchmarkChannelBuffered	39.910 ns/op	new
Windows MSVC	BenchmarkChannelHandoff	2124 ns/op	new
Windows MSVC	BenchmarkDefer	57.570 ns/op	new
Windows MSVC	BenchmarkDirectCall	0.275 ns/op	new
Windows MSVC	BenchmarkGlobalRead	0.350 ns/op	new
Windows MSVC	BenchmarkGlobalWrite	6.969 ns/op	new
Windows MSVC	BenchmarkGoroutine	65059 ns/op	new
Windows MSVC	BenchmarkInterfaceCall	4.489 ns/op	new
Windows MSVC	BenchmarkRuntimeGetG	0.907 ns/op	new

Compared with c0efa083f36f measured in the same runner job. Platforms without a paired baseline are marked new.