0.13.2 --- a thread created from C++ is the thread that is joined

.github/workflows/ci.yml

@@ -593,6 +593,19 @@ jobs:
       # compile. It is run as well as built so that something links afterwards,
       # which is what distinguishes a source that compiles from a package that
       # works.
+      # ⭐ A THREAD CREATED FROM C++ IS THE THREAD THAT IS JOINED.
+      #
+      # musl's C++ `pthread_t` was `unsigned long`, which is thirty-two bits on
+      # Windows, so every std::thread there faulted when joined. The probe's
+      # static_assert is the criterion and the run shows the value surviving a
+      # started context. musl/PATCHES.md, `include/alltypes.h.in`.
+      - name: A thread created from C++ is the thread that is joined
+        env:
+          MCPP_TARGET: ${{ matrix.target }}
+        run: |
+          bash tools/run-probe.sh examples/threads-cxx threads-cxx
+          grep -q 'value 42, result is the argument' examples/threads-cxx/run.log
+
       - name: A program may use the names the internal overlay defines
         env:
           MCPP_TARGET: ${{ matrix.target }}

README.md

@@ -74,7 +74,7 @@ musl reaches its kernel through seven inline functions declared once per
 architecture. Replacing that one header is the whole of the redirection; the
 1345 sources that make up the library are compiled unmodified.
 
-`musl/PATCHES.md` lists the whole of what is not unmodified: **four patched
+`musl/PATCHES.md` lists the whole of what is not unmodified: **five patched
 lines**, all of one kind — a machine word carried through a variable declared
 `long`, which is not a machine word on one of the three targets — and **eleven
 replaced sources**. Five of the eleven are replaced for the same reason: each

examples/threads-cxx/mcpp.toml

@@ -0,0 +1,13 @@
+[package]
+name    = "threads-cxx"
+version = "0.1.0"
+
+[dependencies]
+openkal-musl = { path = "../.." }
+
+[targets.threads-cxx]
+kind = "bin"
+main = "src/main.cpp"
+
+[build]
+cxx_runtime = "host-coupled"

examples/threads-cxx/src/main.cpp

@@ -0,0 +1,42 @@
+/* A thread created from C++ is the thread that is joined.
+ *
+ * musl declares `pthread_t' twice: a pointer for C, and `unsigned long' for C++.
+ * The two agree wherever a long holds a pointer, which is every architecture
+ * musl was written for. Windows is LLP64 --- a long is thirty-two bits --- so a
+ * C++ program kept the lower half of the thread's address, and pthread_join read
+ * through the truncated value. libc++'s std::thread stores exactly this type,
+ * so every std::thread on that system ended in an access violation when joined.
+ *
+ * ⭐ THE static_assert IS THE CRITERION, and it is a compile-time one: before the
+ * change this file does not compile for x86_64-windows-gnu. The run afterwards
+ * shows that the value survives the round trip through a started context.
+ */
+#include <pthread.h>
+#include <stdio.h>
+
+static_assert(sizeof(pthread_t) >= sizeof(void*), "pthread_t must hold the address musl stores in it");
+
+static void* work(void* arg)
+{
+    *static_cast<int*>(arg) = 42;
+    return arg;
+}
+
+int main()
+{
+    int failures = 0;
+    int value = 0;
+    pthread_t thread;
+    if (pthread_create(&thread, nullptr, work, &value) != 0) {
+        puts("pthread_create failed");
+        return 1;
+    }
+    void* result = nullptr;
+    const int joined = pthread_join(thread, &result);
+    printf("joined: %d, value %d, result %s\n", joined, value, result == &value ? "is the argument" : "is not the argument");
+    if (joined != 0) ++failures;
+    if (value != 42) ++failures;
+    if (result != &value) ++failures;
+    printf("-- failures: %d --\n", failures);
+    return failures == 0 ? 0 : 1;
+}

mcpp.toml

@@ -1,7 +1,7 @@
 [package]
 namespace   = "mcpplibs"
 name        = "openkal-musl"
-version     = "0.13.1"
+version     = "0.13.2"
 description = "musl 1.2.5 redirected onto openkal: one C library, ported once, above every implementation of the specification rather than above one kernel."
 license     = "Apache-2.0"
 

musl-generated/aarch64-macos/bits/alltypes.h

@@ -272,7 +272,7 @@ typedef unsigned useconds_t;
 
 #ifdef __cplusplus
 #if defined(__NEED_pthread_t) && !defined(__DEFINED_pthread_t)
-typedef unsigned long pthread_t;
+typedef unsigned _Addr pthread_t;
 #define __DEFINED_pthread_t
 #endif
 

musl-generated/aarch64/bits/alltypes.h

@@ -272,7 +272,7 @@ typedef unsigned useconds_t;
 
 #ifdef __cplusplus
 #if defined(__NEED_pthread_t) && !defined(__DEFINED_pthread_t)
-typedef unsigned long pthread_t;
+typedef unsigned _Addr pthread_t;
 #define __DEFINED_pthread_t
 #endif
 

musl-generated/riscv64/bits/alltypes.h

@@ -262,7 +262,7 @@ typedef unsigned useconds_t;
 
 #ifdef __cplusplus
 #if defined(__NEED_pthread_t) && !defined(__DEFINED_pthread_t)
-typedef unsigned long pthread_t;
+typedef unsigned _Addr pthread_t;
 #define __DEFINED_pthread_t
 #endif
 

musl-generated/x86_64-windows/bits/alltypes.h

@@ -264,7 +264,7 @@ typedef unsigned useconds_t;
 
 #ifdef __cplusplus
 #if defined(__NEED_pthread_t) && !defined(__DEFINED_pthread_t)
-typedef unsigned long pthread_t;
+typedef unsigned _Addr pthread_t;
 #define __DEFINED_pthread_t
 #endif
 

musl-generated/x86_64/bits/alltypes.h

@@ -264,7 +264,7 @@ typedef unsigned useconds_t;
 
 #ifdef __cplusplus
 #if defined(__NEED_pthread_t) && !defined(__DEFINED_pthread_t)
-typedef unsigned long pthread_t;
+typedef unsigned _Addr pthread_t;
 #define __DEFINED_pthread_t
 #endif
 

musl/PATCHES.md

@@ -1,6 +1,6 @@
 # What was changed in musl, and why
 
-musl 1.2.5 is vendored here unmodified except for the four lines below, and the
+musl 1.2.5 is vendored here unmodified except for the five lines below, and the
 list is exhaustive: `git log -p -- musl/` shows every one of them. Each is
 recorded with the reason, because a port that silently edits the library it
 ports is a port nobody can check.
@@ -31,6 +31,28 @@ offset beyond two gigabytes would be truncated, and both would happen silently.
 already widen it because their arguments are wider than a `long`. On every
 target musl supports, the changed line says what the original said.
 
+## `include/alltypes.h.in`, one declaration
+
+```diff
+-TYPEDEF unsigned long pthread_t;
++TYPEDEF unsigned _Addr pthread_t;
+```
+
+musl declares `pthread_t` twice: as a pointer for C, and as an integer for C++,
+because C++ needs a type it can compare and hash. The integer was written
+`unsigned long`, which holds a pointer on every architecture musl supports.
+
+Windows is LLP64, so a C++ program above this library kept the lower half of the
+thread's address. `pthread_join` then read through the truncated value: libc++'s
+`std::thread` stores exactly this type, and every `std::thread` on that system
+ended in an access violation when it was joined (`0xC0000005`, measured on
+windows-2022 and under Wine). `_Addr` is the type musl already uses for
+`size_t`, `uintptr_t` and `ptrdiff_t`; it is `long` on every other target here,
+so the changed line says what the original said there.
+
+The five generated `alltypes.h` headers carry the same one-line change.
+`examples/threads-cxx` does not compile for `x86_64-windows-gnu` without it.
+
 ## `src/stdio/vfwscanf.c`, `src/stdlib/wcstol.c`, `src/stdlib/wcstod.c`
 
 ```diff

musl/include/alltypes.h.in

@@ -51,7 +51,7 @@ TYPEDEF int key_t;
 TYPEDEF unsigned useconds_t;
 
 #ifdef __cplusplus
-TYPEDEF unsigned long pthread_t;
+TYPEDEF unsigned _Addr pthread_t;
 #else
 TYPEDEF struct __pthread * pthread_t;
 #endif