[clang] [clang] Disable C++14 sized deallocation by default for MinGW targets (PR #97232)

[via cfe-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-clang-driver

Author: Martin Storsjö (mstorsjo)

<details>
<summary>Changes</summary>

This reverts 130e93cc26ca9d3ac50ec5a92e3109577ca2e702 for the MinGW target.

This avoids the issue that is discussed in
https://github.com/llvm/llvm-project/issues/96899 (and which is summarized in the code comment). This is intended as a temporary workaround until the issue is handled better within libc++.

---
Full diff: https://github.com/llvm/llvm-project/pull/97232.diff


2 Files Affected:

- (modified) clang/lib/Driver/ToolChains/MinGW.cpp (+22) 
- (modified) clang/unittests/StaticAnalyzer/CallEventTest.cpp (+4) 


``````````diff
diff --git a/clang/lib/Driver/ToolChains/MinGW.cpp b/clang/lib/Driver/ToolChains/MinGW.cpp
index 11e81ebde7eeb..945863a4a8d63 100644
--- a/clang/lib/Driver/ToolChains/MinGW.cpp
+++ b/clang/lib/Driver/ToolChains/MinGW.cpp
@@ -726,6 +726,28 @@ void toolchains::MinGW::addClangTargetOptions(
     }
   }
 
+  // Default to not enabling sized deallocation, but let user provided options
+  // override it.
+  //
+  // If using sized deallocation, user code that invokes delete will end up
+  // calling delete(void*,size_t). If the user wanted to override the
+  // operator delete(void*), there may be a fallback operator delete(void*,size_t)
+  // which calls the regular operator delete(void*).
+  //
+  // However, if the C++ standard library is linked in the form of a DLL,
+  // and the fallback operator delete(void*,size_t) is within this DLL (which is
+  // the case for libc++ at least) it will only redirect towards the library's default
+  // operator delete(void*), not towards the user's provided operator delete(void*).
+  //
+  // This issue can be avoided, if the fallback operators are linked statically
+  // into the callers, even if the C++ standard library is linked as a DLL.
+  //
+  // This is meant as a temporary workaround until libc++ implements this technique,
+  // which is tracked in https://github.com/llvm/llvm-project/issues/96899.
+  if (!DriverArgs.hasArgNoClaim(options::OPT_fsized_deallocation,
+                                options::OPT_fno_sized_deallocation))
+    CC1Args.push_back("-fno-sized-deallocation");
+
   CC1Args.push_back("-fno-use-init-array");
 
   for (auto Opt : {options::OPT_mthreads, options::OPT_mwindows,
diff --git a/clang/unittests/StaticAnalyzer/CallEventTest.cpp b/clang/unittests/StaticAnalyzer/CallEventTest.cpp
index 7c4132788ca7e..ba5b3e4e05ad8 100644
--- a/clang/unittests/StaticAnalyzer/CallEventTest.cpp
+++ b/clang/unittests/StaticAnalyzer/CallEventTest.cpp
@@ -76,7 +76,11 @@ TEST(CXXDeallocatorCall, SimpleDestructor) {
     }
   )",
                                                          Diags));
+#ifdef __MINGW32__
+  EXPECT_EQ(Diags, "test.CXXDeallocator: NumArgs: 1\n");
+#else
   EXPECT_EQ(Diags, "test.CXXDeallocator: NumArgs: 2\n");
+#endif
 }
 
 } // namespace

``````````

</details>


https://github.com/llvm/llvm-project/pull/97232