Fix the round-2 windows walls + port the MSYS2 sweep for the static shape

[ronaldtse]

What this fixes

Two compile walls from tebako-runtime-python PR #16's windows leg (run 34725669811, /tmp log analysis in the commit body):

1. Python/sysmodule.c:3924: 'VPATH' undeclared — the use is #ifdef MS_WINDOWS-gated; MSVC gets VPATH from PC/pyconfig.h, and the autotools sysmodule.o rule passes only -DABIFLAGS/$(MULTIARCH_CPPFLAGS), so the first msys compile of sysmodule.c dies.
2. Python/thread_pthread.h:413: 'native_id' undeclared / :440: implicit declaration of 'pause' — the pthread ladder has no __MINGW32__ rung; upstream's pthread path is POSIX-only and the Windows thread support upstream actually tests is thread_nt.h behind the MSVC build. MSYS2's proven answer is not to grow pthread rungs but to select the NT thread model on mingw.

Then the full sweep: all 103 patches of the MSYS2 MINGW-packages mingw-w64-python series were triaged for our build shape (static-ish, --disable-shared, ucrt64, MS_WINDOWS defined by pyport_ms_windows_msys). Every compile/link-wall patch is ported; every skip is documented below. Two CI rounds after the first push surfaced three more promotions (0005, 0078, 0037) and the syslibs extension — each is called out with its run id.

Ported (MSYS2 number → repo patch → why)

New this round (18):

• 0001 → configure_nt_threads_msys — the thread wall above: --with-nt-threads defaults yes on *-*-mingw*, NT_THREADS defined, pthread/semaphore probes skipped, _POSIX_THREADS undef'd where winpthreads leaks it via <windows.h>. thread.c compiles thread_nt.h; thread_pthread.h never enters the build.
• 0005 → configure_mingw_defaults_msys — promoted from the "cross-only" skip pile on CI evidence (run 34730588522). Despite the MSYS2 title ("default cross configuration"), its configure.ac loader sources Misc/config_mingw for EVERY *-*-mingw* host, native included — and that file pins exactly the probes the first CI leg died on: ac_cv_func_alarm=ignore (the link probe finds alarm in the mingw import libs, but ucrt64's <unistd.h> declares it only under __USE_MINGW_ALARM → signalmodule.c:398 implicit declaration; MSVC has no HAVE_ALARM), ac_cv_file__dev_ptmx=ignore + ac_cv_file__dev_ptc=no (the MSYS2 shell emulates /dev/ptmx, so the native probe defines HAVE_DEV_PTMX and os_openpty_impl reaches the STREAMS ioctl(I_PUSH) calls mingw-w64 lacks → posixmodule.c:9000; os.openpty simply does not exist on Windows, matching MSVC), ftruncate/truncate pinned off (wrong mingw semantics; the _chsize path is the Windows-correct one), getpeername forced on. Misc/cross_mingw32 rides along verbatim (inert natively).
• 0008 → posixmodule_nt_msys — the module must be named nt on mingw (PyInit_nt vs the posix registration; os.py imports nt when MACHDEP=win32). Boot/link wall.
• 0016 → frozenmain_drop_msys — frozenmain.o references PC/frozen_dllmain.c symbols → exe link wall.
• 0017 → socketmodule_msys — without it _socket never builds on mingw (header gates, SOCKET_LIBS, ws2tcpip probes).
• 0036 → signal_ws2_32_msys — signalmodule.c:773 calls WSAGetLastError() under MS_WINDOWS; _signal is a bootstrap module so its .o rides MODULE_OBJS into the _bootstrap_python/_freeze_module/exe links, which carry $(LIBS) $(MODLIBS) $(SYSLIBS) but never the factory's TEBAKO_LIBS. The trailing -lws2_32 folds into BASEMODLIBS → MODLIBS (makesetup), a var the factory's modlib_rewrites never touches. Applies after 0008 (its context line @MODULE_POSIXMODULE_NAME@ is 0008's output).
• 0037 → io_winconsoleio_msys — promoted from the "feature parity" skip pile on CI evidence (run 34732802820 — a LINK wall this time). Upstream's autotools build never compiles Modules/_io/winconsoleio.c (the MSVC vcxproj carries it instead), so once 0002's HAVE_WINDOWS_CONSOLE_IO arm makes the console code paths live on mingw, the _freeze_module link dies on _get_console_type (Parser/myreadline.c:273/:276), _PyIO_get_console_type and winconsoleio_spec (Modules/_io/_iomodule.c:324/:584). The patch appends _io/winconsoleio.c to the _io bootstrap line (the file's whole body is #ifdef HAVE_WINDOWS_CONSOLE_IO — POSIX builds gain an empty object) and registers _testconsole on win32 (test-only, rides along verbatim).
• 0041 → configure_inet_pton_msys — 0017's probe pair.
• 0045 → timemodule_msys — ucrt _timezone spellings (strict-ANSI risk).
• 0054 → sysmodule_vpath_msys — the VPATH wall above.
• 0056 → bootstrap_python_link_flags_msys — _bootstrap_python defines wmain; mingw's CRT only calls it under -municode → freeze-tool link wall on every build.
• 0059 → ssl_module_msys — static openssl needs -lws2_32 -lcrypt32; plus the cert-store widening.
• 0060 → configure_winsock_netdb_msys — promoted on cross-probe evidence. Without it every netdb probe fails on mingw (PY_CHECK_FUNC compiles void *x=<fn> against <netdb.h>, which on mingw-w64 declares none of them), leaving HAVE_GETHOSTNAME/HAVE_SHUTDOWN undefined while MSVC's PC/pyconfig.h defines both — so the 0017-enabled _socket would lose socket.gethostname() (#ifdef HAVE_GETHOSTNAME, socketmodule.c:5863) and shutdown versus the Windows reference. Verified by this PR's cross probe (before: all netdb HAVE_ undef; after: both defined). HAVE_GETADDRINFO/HAVE_GETNAMEINFO stay undefined exactly as upstream's PC build has them.
• 0063 → configure_exeext_msys — deterministic .exe (upstream: empty EXEEXT + a case-probe).
• 0075 → configure_disable_posix_modules_msys — the POSIX-only module set is n/a on mingw (subsumes 0058).
• 0078 → configure_noerror_incompat_ptr_msys — promoted from the "warning-grade" skip pile: GCC 14 escalated -Wincompatible-pointer-types from warning to ERROR (CI ships gcc 16). The MSVC-oriented MS_WINDOWS branches of posixmodule.c pass wchar_t** where the CRT prototypes take const wchar_t* const* (PathCchSkipRoot :5269, _wexecv/_wexecve :7062/:7150, _wspawnv/_wspawnve :7823/:7945) — upstream code MSVC accepts. MSYS2's answer is the blanket demotion on mingw hosts rather than diverging from upstream at every call site (the qualifiers are ABI-honored either way).
• 0097 → mmapmodule_no_seh_msys — the mmap gate passes on mingw (sys/mman.h exists), so the __try/__except SEH code hits GCC → compile error.
• 0100 → configure_pthread_name_maxlen_msys — _PYTHREAD_NAME_MAXLEN is used UNGUARDED in _threadmodule.c's Windows branch → compile wall.

Already in the tree from the earlier rounds (5): 0002 (pyport_ms_windows_msys), 0004 (configure_machdep_msys), 0007 (configure_pc_include_msys), 0052 (configure_c_locale_coercion_msys), 0096 (pycore_time_timeval_msys).

Plus one tebako deviation (the shape is ours; the lib set is MSYS2's):

• configure_syslibs_msys — the factory's Mlibs.modlib_rewrites rewrites the generated Makefile's MODULE__SSL_LDFLAGS/MODULE__HASHLIB_LDFLAGS (and zlib/binascii) to absolute .a paths, dropping the -lws2_32 -lcrypt32 tail 0059 appends. Every link pulling _ssl.o/_hashlib.o then dies on undefined WSAStartup/CertOpenSystemStore/BCryptGenRandom. MSYS2 never needs the indirection (shared build: libpython3.dll resolves the tails through its own import list). Fix source-side: a host-conditional WINDOWS_SYSLIBS rides SYSLIBS, which sits after MODLIBS on every link line and is never rewritten downstream. Extended at run 34732802820 with the -lversion -lshlwapi -lpathcch tail of 0010's mingw LIBS hunk (0010 itself stays skipped as shared-build machinery): the same run's _freeze_module link died on GetFileVersionInfoSizeW/GetFileVersionInfoW/VerQueryValueW (Python/sysmodule.c:1666-1674 → -lversion) and __imp_PathCchSkipRoot/__imp_PathCchCombineEx (Python/fileutils.c:2160/:2407, Modules/posixmodule.c:5269 → -lpathcch; pathconfig.c's <shlwapi.h> arm → -lshlwapi). MSVC gets those libs from #pragma comment(lib, ...) in the sources — which GCC silently ignores. Full set: -lws2_32 -lcrypt32 -ladvapi32 -lbcrypt -lversion -lshlwapi -lpathcch on *-*-mingw*, empty elsewhere.

All generated-configure/pyconfig.h.in hunks are machine-exact regens with the upstream-pinned autoconf 2.72, not hand-syncs — proven by regenerating over the assembled 24-patch tree (configure/pyconfig.h.in come out byte-identical). The full series applied to pristine 3.14.7 in manifest order reproduces an independently staged sequential-MSYS2 reference tree byte-for-byte.

Skipped (every other patch in the 103, grouped)

• Feature parity, not compile walls (deferred; flagged for a follow-up runtime-parity PR): 0018 (enables winreg/msvcrt/_winapi/winsound/overlapped — note _winapi is subprocess-critical on Windows, first candidate for the next round), 0038/0061/0073 (multiprocessing set), 0003 (sys flags), 0012 + 0066–0070 (dynload_win set — ucrt64 has dlfcn/dlopen; the VFS-path LoadLibrary question is factory/contract scope), 0062 (mmap enablement — the gate already passes), 0074 (selectmodule — examined this round: no compile wall (the POSIX branch compiles on mingw), but without it select() on sockets takes the _PyIsSelectable_fd range-check branch MSVC never runs → runtime divergence; joins the parity batch), 0042 (frozen importlib path-separator order — runtime behavior), 0043 (mingw stack reserve — link-flag parity with MSVC's bpo-40458 fix; no wall), 0050 (clock probes — MSVC-feature parity / winpthread dependency hygiene; in our shape the probes succeed AND link cleanly against default ucrt64 libs, cross-probe verified — if a winpthread undefined-ref ever appears at the factory, this is the queued fix), 0065 (pathcch forward-slash normalization — runtime correctness on / paths, not a wall; also a parity-batch candidate). (0037 was in this bucket in the first pass — promoted on link-wall evidence, see above.)
• Shared-build / DLL naming / packaging (we are --disable-shared): 0009, 0010 (its mingw LIBS hunk is folded into the syslibs deviation — see above), 0011, 0013, 0014, 0015, 0028, 0029, 0034, 0039, 0047, 0048, 0049, 0071, 0086, 0088, 0090, 0093, 0098 (ABIFLAGS is empty in our shape), 0103.
• Behavior / install-path, non-wall (portable fallbacks verified to exist): 0019–0022, 0026, 0027, 0030, 0035, 0040, 0057 (getpath — PYTHONHOME drives runtime), 0083, 0085, 0091, 0099.
• Warning-grade only (no -Werror on these classes): 0031, 0033, 0081, 0082. (0078 was in this bucket in the first pass — promoted, see above.)
• Cross-build only (we build natively in the ucrt64 shell): 0055, 0080. (0005 was mis-filed here in the first pass — its config_mingw loads natively; promoted, see above.)
• 32-bit / debug / policy: 0006, 0072, 0092, 0094 (_WIN32_WINNT floor).
• Tests/CI: 0044, 0051, 0053, 0079, 0084, 0087, 0095, 0102.
• Inapplicable in our shape: 0023–0025 + 0076 (_ctypes disabled via the factory's Setup.local), 0077 (_uuid disabled), 0089 (_wmi not in the autoconf module set at all), 0032 (gettext absent in the CI env), 0046/0064 (dlopen probes harmless — mingw-w64 ships dlfcn), 0058 (subsumed by 0075), 0101 (autoconf version pin — we regenerate with the pinned 2.72 ourselves).

Tooling changes the sweep forced

• Tfs::SourcePrep#audit now applies the series cumulatively in manifest order instead of git apply --check per patch on pristine. A quilt series cannot satisfy per-patch-pristine (0036's context line @MODULE_POSIXMODULE_NAME@ only exists after 0008). On a mid-series failure the remaining patches report a not-checked failure. Specs pin the dependent-context and not-checked-tail behaviors (103 examples, +2); the lint/README/onboarder comments follow.
• tools/compile_smoke WALL_OBJECTS gains the round-2 walls and the patched module TUs per the regression-list contract: Python/sysmodule.o, Python/thread.o, Modules/posixmodule.o, Modules/signalmodule.o, Modules/timemodule.o, Modules/socketmodule.o, Modules/_threadmodule.o, Programs/_bootstrap_python.o — plus a platform-conditional WALL_OBJECTS_MSYS for Modules/_io/winconsoleio.o: the TU exists only in the patched tree (0037 adds it to the _io bootstrap line), so the pristine POSIX leg has no rule for it and must not fail on its absence, while on windows-msys a missing rule IS the regression. Patch-named .c auto-adds now probe the generated Makefile for a literal rule first: an n/a module (e.g. _ssl without an openssl dev package in the smoke env) emits NO rule, and make's implicit %.o:%.c fallback would otherwise compile it with the wrong flags and fail. Skips are reported in the OK line.

Verification (local; the native ucrt64 compile is CI's oracle)

• tools/lint 3.14.7: 24-patch series applies clean to the pristine tarball (cumulative audit).
• Series output tree byte-identical to the independently staged manifest-order reference; shipped configure/pyconfig.h.in equal the pinned autoconf 2.72 regen of the shipped configure.ac exactly.
• rspec 103/0, tools/validate_manifests OK.
• POSIX smoke leg: tools/compile_smoke 3.14.7 --platform linux-gnu — all 12 wall TUs compile on the pristine tree (regression proof for the POSIX asset).
• Cross configure (x86_64-w64-mingw32-gcc 14, --host override, --disable-shared, NO manual ac_cv presets — config_mingw loads itself): NT_THREADS defined with _POSIX_THREADS undef'd; nt posixmodule.c and _signal signalmodule.c -lws2_32 in Setup.bootstrap; EXE/BUILDEXE=.exe; MODULE__SOCKET_STATE=yes with the ws2_32 tail; HAVE_GETHOSTNAME/HAVE_SHUTDOWN defined (MSVC parity); HAVE_ALARM/HAVE_DEV_PTMX/HAVE_TRUNCATE/HAVE_FTRUNCATE pinned off; HAVE_GETPEERNAME on; _posixshmem/grp/resource n/a; _PYTHREAD_NAME_MAXLEN=32766; the extended SYSLIBS tail; -Wno-error=incompatible-pointer-types in CFLAGS_NODIST; -DVPATH + -I./PC on the make -n Python/sysmodule.o line.
• Cross compile+link sweep: all 13 wall TUs compile clean; the full core (libpython3.14.a, 191 objects) cross-compiles with zero errors; and Programs/_freeze_module — the run 34732802820 casualty — now cross-LINKS clean (winconsoleio.o in the _io objects; version/pathcch resolved by the extended SYSLIBS).

CI history: run 34730588522 failed on exactly the two probe walls 0005 pins (alarm, /dev/ptmx) plus the _bootstrap_python.o cascade from them; the earlier walls (sysmodule/thread) compiled. Run 34732802820 got through every compile and died at the _freeze_module LINK on the console symbols (0037) and the version/pathcch libs (the syslibs extension). The 0078 const-qualification errors were found by the local cross compile before the second push.