[libc++] Fix ABI break introduced by switching to _LIBCPP_COMPRESSED_PAIR #154686

philnik777 · 2025-08-21T07:26:48Z

LLVM 20 contained an ABI break that can result in the size of std::unordered_{map,set,multimap,multiset} and std::deque changing when used with an allocator type that is empty and contains a base class that is the same across rebound allocator instantiations (e.g. Allocator<int> and Allocator<char> are both empty and contain the same base class).

In addition, the layout of a user-defined type that:
- contains one of the following containers: std::unordered_{map,set,multimap,multiset}, std::deque, std::map, std::set, std::multimap, std::multiset, std::list or std::vector, and
- passes an empty allocator, comparator or hasher type to that container, and
- has a member of that same empty allocator, comparator or hasher type inside the enclosing struct, and
- that member is either marked with [[no_unique_address]] or optimized out via the EBO (empty base optimization) technique
saw its size increase from LLVM 19 to LLVM 20. This was caused by the usage of [[no_unique_address]] within some of libc++'s containers in a way that allowed subtle interactions with enclosing objects. This is fixed in LLVM 21 on Clang (returning to the LLVM 19 ABI), however that implies an ABI break from LLVM 20 to LLVM 21.
Furthermore, fixing this causes a slight regression to constant evaluation support in std::unique_ptr. Specifically, constant evaluation will now fail when the deleter relies on being value-initialized for constant-evaluation admissibility. If a default-initialized deleter can be used during constant evaluation, or if the default constructor is non-trivial, the unique_ptr is not affected by this regression. In particular, this regression does not impact any unique_ptr using the default deleter.

Note that there is currently no way to realistically fix this ABI break on GCC, therefore GCC will remain on the ABI introduced in LLVM 19. That also means that Clang and GCC will have a slightly different ABI for the small subset of types listed above until we are able to apply the same fix we did with Clang on GCC.

We fix this regression by surrounding the members of the _LIBCPP_COMPRESSED_PAIR with an anonymous struct. This restricts the shifting of empty types to the front of the _LIBCPP_COMPRESSED_PAIR instead of throughout the surrounding object. This "frees up" the zero offset to contain another object of the same type, restoring the ability to perform EBO or to elide the storage for a type with [[no_unique_address]] in the enclosing (user-defined) struct.

Fixes #154146

github-actions · 2025-08-21T07:29:04Z

✅ With the latest revision this PR passed the C/C++ code formatter.

ldionne · 2025-08-21T14:38:51Z

libcxx/include/__memory/compressed_pair.h

I don't think the attribute is needed anymore now that we have a struct enclosing it. However, I wouldn't make that change now, I would do it in a followup patch since that wouldn't be an ABI change, just a simplification.

libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/default.pass.cpp

libcxx/include/__memory/compressed_pair.h

llvmbot · 2025-08-22T13:28:07Z

@llvm/pr-subscribers-libcxx

Author: Nikolas Klauser (philnik777)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/154686.diff

14 Files Affected:

(modified) libcxx/docs/ReleaseNotes/21.rst (+14)
(modified) libcxx/include/__memory/compressed_pair.h (+38-15)
(modified) libcxx/include/string (+12-2)
(modified) libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp (+1-3)
(modified) libcxx/test/libcxx/containers/associative/set/abi.compile.pass.cpp (+1-3)
(modified) libcxx/test/libcxx/containers/associative/unord.map/abi.compile.pass.cpp (+2-4)
(modified) libcxx/test/libcxx/containers/associative/unord.set/abi.compile.pass.cpp (+2-4)
(modified) libcxx/test/libcxx/containers/sequences/deque/abi.compile.pass.cpp (+2-4)
(modified) libcxx/test/libcxx/containers/sequences/list/abi.compile.pass.cpp (+1-3)
(modified) libcxx/test/libcxx/containers/sequences/vector.bool/abi.compile.pass.cpp (+1-3)
(modified) libcxx/test/libcxx/containers/sequences/vector/abi.compile.pass.cpp (+1-3)
(modified) libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/default.pass.cpp (+4)
(modified) libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/nullptr.pass.cpp (+4)
(modified) libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/pointer.pass.cpp (+4)

diff --git a/libcxx/docs/ReleaseNotes/21.rst b/libcxx/docs/ReleaseNotes/21.rst
index 91123ffa3e34b..1eaba67b18465 100644
--- a/libcxx/docs/ReleaseNotes/21.rst
+++ b/libcxx/docs/ReleaseNotes/21.rst
@@ -148,6 +148,20 @@ ABI Affecting Changes
   comparison between shared libraries, since all RTTI has the correct visibility now. There is no behaviour change on
   Clang.
 
+- LLVM 20 contains an ABI break when
+  (1) using empty allocators which derive from the same base class, or
+  (2) having an object of the same type as an empty member in one of the following classes that could be layed out at
+      the beginnig of the object
+
+  (1) applies to ``unordered_map``, ``unordered_set``, ``unordered_multimap``, ``unordered_multiset``  and ``deque``.
+  (2) applies to the containers listed in (1) as well as ``map``, ``set``, ``multimap``, ``multiset``, ``list`` and
+      ``vector``.
+
+  Fixing this causes a regression in ``unique_ptr``. Specifically, constant evaluation fails when the deleter relies on
+  being value-initialized. If the deleter works if it is default initialized or is not trivially default constructible
+  it is not affected.
+
+  For more details see https://llvm.org/PR154146.
 
 Build System Changes
 --------------------
diff --git a/libcxx/include/__memory/compressed_pair.h b/libcxx/include/__memory/compressed_pair.h
index fb7b7b7afcc8c..220dafe856716 100644
--- a/libcxx/include/__memory/compressed_pair.h
+++ b/libcxx/include/__memory/compressed_pair.h
@@ -80,21 +80,44 @@ class __compressed_pair_padding {
 template <class _ToPad>
 class __compressed_pair_padding<_ToPad, true> {};
 
-#  define _LIBCPP_COMPRESSED_PAIR(T1, Initializer1, T2, Initializer2)                                                  \
-    _LIBCPP_NO_UNIQUE_ADDRESS __attribute__((__aligned__(::std::__compressed_pair_alignment<T2>))) T1 Initializer1;    \
-    _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T1> _LIBCPP_CONCAT3(__padding1_, __LINE__, _);          \
-    _LIBCPP_NO_UNIQUE_ADDRESS T2 Initializer2;                                                                         \
-    _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T2> _LIBCPP_CONCAT3(__padding2_, __LINE__, _)
-
-#  define _LIBCPP_COMPRESSED_TRIPLE(T1, Initializer1, T2, Initializer2, T3, Initializer3)                              \
-    _LIBCPP_NO_UNIQUE_ADDRESS                                                                                          \
-    __attribute__((__aligned__(::std::__compressed_pair_alignment<T2>),                                                \
-                   __aligned__(::std::__compressed_pair_alignment<T3>))) T1 Initializer1;                              \
-    _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T1> _LIBCPP_CONCAT3(__padding1_, __LINE__, _);          \
-    _LIBCPP_NO_UNIQUE_ADDRESS T2 Initializer2;                                                                         \
-    _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T2> _LIBCPP_CONCAT3(__padding2_, __LINE__, _);          \
-    _LIBCPP_NO_UNIQUE_ADDRESS T3 Initializer3;                                                                         \
-    _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T3> _LIBCPP_CONCAT3(__padding3_, __LINE__, _)
+#  ifdef _LIBCPP_COMPILER_GCC
+#    define _LIBCPP_COMPRESSED_PAIR(T1, Initializer1, T2, Initializer2)                                                \
+      _LIBCPP_NO_UNIQUE_ADDRESS __attribute__((__aligned__(::std::__compressed_pair_alignment<T2>))) T1 Initializer1;  \
+      _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T1> _LIBCPP_CONCAT3(__padding1_, __LINE__, _);        \
+      _LIBCPP_NO_UNIQUE_ADDRESS T2 Initializer2;                                                                       \
+      _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T2> _LIBCPP_CONCAT3(__padding2_, __LINE__, _)
+
+#    define _LIBCPP_COMPRESSED_TRIPLE(T1, Initializer1, T2, Initializer2, T3, Initializer3)                            \
+      _LIBCPP_NO_UNIQUE_ADDRESS                                                                                        \
+      __attribute__((__aligned__(::std::__compressed_pair_alignment<T2>),                                              \
+                     __aligned__(::std::__compressed_pair_alignment<T3>))) T1 Initializer1;                            \
+      _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T1> _LIBCPP_CONCAT3(__padding1_, __LINE__, _);        \
+      _LIBCPP_NO_UNIQUE_ADDRESS T2 Initializer2;                                                                       \
+      _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T2> _LIBCPP_CONCAT3(__padding2_, __LINE__, _);        \
+      _LIBCPP_NO_UNIQUE_ADDRESS T3 Initializer3;                                                                       \
+      _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T3> _LIBCPP_CONCAT3(__padding3_, __LINE__, _)
+#  else
+#    define _LIBCPP_COMPRESSED_PAIR(T1, Initializer1, T2, Initializer2)                                                \
+      struct {                                                                                                         \
+        _LIBCPP_NO_UNIQUE_ADDRESS                                                                                      \
+        __attribute__((__aligned__(::std::__compressed_pair_alignment<T2>))) T1 Initializer1;                          \
+        _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T1> _LIBCPP_CONCAT3(__padding1_, __LINE__, _);      \
+        _LIBCPP_NO_UNIQUE_ADDRESS T2 Initializer2;                                                                     \
+        _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T2> _LIBCPP_CONCAT3(__padding2_, __LINE__, _);      \
+      }
+
+#    define _LIBCPP_COMPRESSED_TRIPLE(T1, Initializer1, T2, Initializer2, T3, Initializer3)                            \
+      struct {                                                                                                         \
+        _LIBCPP_NO_UNIQUE_ADDRESS                                                                                      \
+        __attribute__((__aligned__(::std::__compressed_pair_alignment<T2>),                                            \
+                       __aligned__(::std::__compressed_pair_alignment<T3>))) T1 Initializer1;                          \
+        _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T1> _LIBCPP_CONCAT3(__padding1_, __LINE__, _);      \
+        _LIBCPP_NO_UNIQUE_ADDRESS T2 Initializer2;                                                                     \
+        _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T2> _LIBCPP_CONCAT3(__padding2_, __LINE__, _);      \
+        _LIBCPP_NO_UNIQUE_ADDRESS T3 Initializer3;                                                                     \
+        _LIBCPP_NO_UNIQUE_ADDRESS ::std::__compressed_pair_padding<T3> _LIBCPP_CONCAT3(__padding3_, __LINE__, _);      \
+      }
+#  endif
 
 #else
 #  define _LIBCPP_COMPRESSED_PAIR(T1, Name1, T2, Name2)                                                                \
diff --git a/libcxx/include/string b/libcxx/include/string
index 3c1523eb5de33..1d197654b9fee 100644
--- a/libcxx/include/string
+++ b/libcxx/include/string
@@ -974,7 +974,12 @@ public:
 
   _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 basic_string()
       _NOEXCEPT_(is_nothrow_default_constructible<allocator_type>::value)
-      : __rep_() {
+#  if _LIBCPP_STD_VER >= 20 // TODO(LLVM 23): Remove this condition; this is a workaround for https://llvm.org/PR154567
+      : __rep_(__short())
+#  else
+      : __rep_()
+#  endif
+  {
     __annotate_new(0);
   }
 
@@ -984,7 +989,12 @@ public:
 #  else
       _NOEXCEPT
 #  endif
-      : __rep_(), __alloc_(__a) {
+#  if _LIBCPP_STD_VER >= 20 // TODO(LLVM 23): Remove this condition; this is a workaround for https://llvm.org/PR154567
+      : __rep_(__short()),
+#  else
+      : __rep_(),
+#  endif
+        __alloc_(__a) {
     __annotate_new(0);
   }
 
diff --git a/libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp
index 04c1802bc84f6..908338631bb51 100644
--- a/libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp
@@ -8,8 +8,6 @@
 
 // UNSUPPORTED: libcpp-abi-no-compressed-pair-padding
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <map>
 
@@ -90,7 +88,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 32, "");
+static_assert(sizeof(user_struct) == 24, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(map_alloc<int, std::allocator<std::pair<const int, int> > >) == 24, "");
diff --git a/libcxx/test/libcxx/containers/associative/set/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/associative/set/abi.compile.pass.cpp
index 1b85bb7c93773..626f573967efb 100644
--- a/libcxx/test/libcxx/containers/associative/set/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/associative/set/abi.compile.pass.cpp
@@ -8,8 +8,6 @@
 
 // UNSUPPORTED: libcpp-abi-no-compressed-pair-padding
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <set>
 
@@ -90,7 +88,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 32, "");
+static_assert(sizeof(user_struct) == 24, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(set_alloc<int, std::allocator<int> >) == 24, "");
diff --git a/libcxx/test/libcxx/containers/associative/unord.map/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/associative/unord.map/abi.compile.pass.cpp
index 0476872487036..e6d27d245c472 100644
--- a/libcxx/test/libcxx/containers/associative/unord.map/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/associative/unord.map/abi.compile.pass.cpp
@@ -12,8 +12,6 @@
 // unordered containers changes when bounded unique_ptr is enabled.
 // UNSUPPORTED: libcpp-has-abi-bounded-unique_ptr
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <unordered_map>
 
@@ -94,7 +92,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 48, "");
+static_assert(sizeof(user_struct) == 40, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(unordered_map_alloc<int, std::allocator<std::pair<const int, int> > >) == 40, "");
@@ -126,7 +124,7 @@ struct TEST_ALIGNAS(32) AlignedHash {};
 struct UnalignedEqualTo {};
 
 // This part of the ABI has been broken between LLVM 19 and LLVM 20.
-static_assert(sizeof(std::unordered_map<int, int, AlignedHash, UnalignedEqualTo>) == 64, "");
+static_assert(sizeof(std::unordered_map<int, int, AlignedHash, UnalignedEqualTo>) == 96, "");
 static_assert(TEST_ALIGNOF(std::unordered_map<int, int, AlignedHash, UnalignedEqualTo>) == 32, "");
 
 #elif __SIZE_WIDTH__ == 32
diff --git a/libcxx/test/libcxx/containers/associative/unord.set/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/associative/unord.set/abi.compile.pass.cpp
index e9d88a126de6c..1658e7e33f850 100644
--- a/libcxx/test/libcxx/containers/associative/unord.set/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/associative/unord.set/abi.compile.pass.cpp
@@ -12,8 +12,6 @@
 // unordered containers changes when bounded unique_ptr is enabled.
 // UNSUPPORTED: libcpp-has-abi-bounded-unique_ptr
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <unordered_set>
 
@@ -94,7 +92,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 48, "");
+static_assert(sizeof(user_struct) == 40, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(unordered_set_alloc<int, std::allocator<int> >) == 40, "");
@@ -125,7 +123,7 @@ struct TEST_ALIGNAS(32) AlignedHash {};
 struct UnalignedEqualTo {};
 
 // This part of the ABI has been broken between LLVM 19 and LLVM 20.
-static_assert(sizeof(std::unordered_set<int, AlignedHash, UnalignedEqualTo>) == 64, "");
+static_assert(sizeof(std::unordered_set<int, AlignedHash, UnalignedEqualTo>) == 96, "");
 static_assert(TEST_ALIGNOF(std::unordered_set<int, AlignedHash, UnalignedEqualTo>) == 32, "");
 
 #elif __SIZE_WIDTH__ == 32
diff --git a/libcxx/test/libcxx/containers/sequences/deque/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/sequences/deque/abi.compile.pass.cpp
index 38f957f416c87..a19897cbb1588 100644
--- a/libcxx/test/libcxx/containers/sequences/deque/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/sequences/deque/abi.compile.pass.cpp
@@ -8,8 +8,6 @@
 
 // UNSUPPORTED: libcpp-abi-no-compressed-pair-padding
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <deque>
 
@@ -88,14 +86,14 @@ static_assert(sizeof(std::deque<int, min_allocator<int> >) == 48, "");
 static_assert(sizeof(std::deque<int, test_allocator<int> >) == 80, "");
 static_assert(sizeof(std::deque<int, small_iter_allocator<int> >) == 12, "");
 static_assert(sizeof(std::deque<int, final_small_iter_allocator<int> >) == 16, "");
-static_assert(sizeof(std::deque<int, common_base_allocator<int> >) == 56, "");
+static_assert(sizeof(std::deque<int, common_base_allocator<int> >) == 48, "");
 
 static_assert(sizeof(std::deque<char>) == 48, "");
 static_assert(sizeof(std::deque<char, min_allocator<char> >) == 48, "");
 static_assert(sizeof(std::deque<char, test_allocator<char> >) == 80, "");
 static_assert(sizeof(std::deque<char, small_iter_allocator<char> >) == 12, "");
 static_assert(sizeof(std::deque<char, final_small_iter_allocator<char> >) == 16, "");
-static_assert(sizeof(std::deque<char, common_base_allocator<char> >) == 56, "");
+static_assert(sizeof(std::deque<char, common_base_allocator<char> >) == 48, "");
 
 static_assert(TEST_ALIGNOF(std::deque<int>) == 8, "");
 static_assert(TEST_ALIGNOF(std::deque<int, min_allocator<int> >) == 8, "");
diff --git a/libcxx/test/libcxx/containers/sequences/list/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/sequences/list/abi.compile.pass.cpp
index 1737497ab159e..ad10051e514fe 100644
--- a/libcxx/test/libcxx/containers/sequences/list/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/sequences/list/abi.compile.pass.cpp
@@ -6,8 +6,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <list>
 
@@ -65,7 +63,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 32, "");
+static_assert(sizeof(user_struct) == 24, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(std::list<int>) == 24, "");
diff --git a/libcxx/test/libcxx/containers/sequences/vector.bool/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/sequences/vector.bool/abi.compile.pass.cpp
index 97e09ab4ca868..109e97a3fe8be 100644
--- a/libcxx/test/libcxx/containers/sequences/vector.bool/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/sequences/vector.bool/abi.compile.pass.cpp
@@ -8,8 +8,6 @@
 
 // UNSUPPORTED: libcpp-abi-no-compressed-pair-padding
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <vector>
 
@@ -67,7 +65,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 32, "");
+static_assert(sizeof(user_struct) == 24, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(std::vector<bool>) == 24, "");
diff --git a/libcxx/test/libcxx/containers/sequences/vector/abi.compile.pass.cpp b/libcxx/test/libcxx/containers/sequences/vector/abi.compile.pass.cpp
index 879e143510752..1a55d1359eeb0 100644
--- a/libcxx/test/libcxx/containers/sequences/vector/abi.compile.pass.cpp
+++ b/libcxx/test/libcxx/containers/sequences/vector/abi.compile.pass.cpp
@@ -6,8 +6,6 @@
 //
 //===----------------------------------------------------------------------===//
 
-// XFAIL: FROZEN-CXX03-HEADERS-FIXME
-
 #include <cstdint>
 #include <vector>
 
@@ -73,7 +71,7 @@ struct user_struct {
 };
 
 #if __SIZE_WIDTH__ == 64
-static_assert(sizeof(user_struct) == 32, "");
+static_assert(sizeof(user_struct) == 24, "");
 static_assert(TEST_ALIGNOF(user_struct) == 8, "");
 
 static_assert(sizeof(std::vector<int>) == 24, "");
diff --git a/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/default.pass.cpp b/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/default.pass.cpp
index 7f4c90922d6c3..e2abfd17d27c3 100644
--- a/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/default.pass.cpp
+++ b/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/default.pass.cpp
@@ -82,6 +82,10 @@ TEST_CONSTEXPR_CXX23 bool test_basic() {
     p.get_deleter().set_state(5);
     assert(p.get_deleter().state() == 5);
   }
+// TODO: Remove this check once https://llvm.org/PR154567 is fixed
+#if TEST_STD_VER >= 23 && defined(TEST_COMPILER_CLANG)
+  if (!TEST_IS_CONSTANT_EVALUATED)
+#endif
   {
     std::unique_ptr<ElemType, DefaultCtorDeleter<ElemType> > p;
     assert(p.get() == 0);
diff --git a/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/nullptr.pass.cpp b/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/nullptr.pass.cpp
index 45017a03b95dd..f5e0541684d13 100644
--- a/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/nullptr.pass.cpp
+++ b/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/nullptr.pass.cpp
@@ -47,6 +47,10 @@ TEST_CONSTEXPR_CXX23 void test_basic() {
     assert(p.get() == 0);
     assert(p.get_deleter().state() == 0);
   }
+// TODO: Remove this check once https://llvm.org/PR154567 is fixed
+#if TEST_STD_VER >= 23 && defined(TEST_COMPILER_CLANG)
+  if (!TEST_IS_CONSTANT_EVALUATED)
+#endif
   {
     std::unique_ptr<VT, DefaultCtorDeleter<VT> > p(nullptr);
     assert(p.get() == 0);
diff --git a/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/pointer.pass.cpp b/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/pointer.pass.cpp
index cbce5c9c74c5a..e9912d4574af6 100644
--- a/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/pointer.pass.cpp
+++ b/libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/pointer.pass.cpp
@@ -75,6 +75,10 @@ TEST_CONSTEXPR_CXX23 void test_pointer() {
   }
   if (!TEST_IS_CONSTANT_EVALUATED)
     assert(A::count == 0);
+// TODO: Remove this check once https://llvm.org/PR154567 is fixed
+#if TEST_STD_VER >= 23 && defined(TEST_COMPILER_CLANG)
+  if (!TEST_IS_CONSTANT_EVALUATED)
+#endif
   {
     A* p = newValue<ValueT>(expect_alive);
     if (!TEST_IS_CONSTANT_EVALUATED)

libcxx/docs/ReleaseNotes/21.rst

ldionne · 2025-08-22T13:48:46Z

libcxx/test/libcxx/containers/associative/unord.map/abi.compile.pass.cpp

Is that still the case? I thought we were fixing this (accepted) ABI break at the same time?

libcxx/test/libcxx/containers/associative/unord.set/abi.compile.pass.cpp

libcxx/include/__memory/compressed_pair.h

libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp

ldionne

Thanks a lot for fixing this. LGTM once CI is green. Once this is landed, we'll also cherry-pick it to the LLVM 21 release.

philnik777 · 2025-08-22T18:32:01Z

@Michael137 Are the lldb failures related? (Even if they are I'll land this, since it's very time sensitive, just want to know whether there is follow up work to do)

…PAIR Update libcxx/docs/ReleaseNotes/21.rst Co-authored-by: Louis Dionne <[email protected]>

h-vetinari · 2025-08-23T06:17:24Z

libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.class/unique.ptr.ctor/nullptr.pass.cpp

+// TODO: Remove this check once https://llvm.org/PR154567 is fixed
+#if TEST_STD_VER >= 23 && defined(TEST_COMPILER_CLANG)
+  if (!TEST_IS_CONSTANT_EVALUATED)
+#endif


#154567 has been fixed already. Is this still needed/desired?

Yes, we support older compilers.

I meant that this deviates from the // TODO(LLVM 23): format elsewhere in this PR, and wasn't clear to me why, hence the question

boomanaiden154 · 2025-08-23T18:56:04Z

The LLDB formatter failures are showing up in premerge. It would be nice if they could be fixed soonish.

Michael137 · 2025-08-23T19:03:57Z

The LLDB formatter failures are showing up in premerge. It would be nice if they could be fixed soonish.

I'll only be able to take a look evening UK time today at the earliest

slydiman · 2025-08-24T07:00:13Z

Note the following buildbots are broken

https://lab.llvm.org/buildbot/#/builders/195/builds/13641
https://lab.llvm.org/buildbot/#/builders/197/builds/8319

Michael137 · 2025-08-24T08:32:49Z

Note the following buildbots are broken

https://lab.llvm.org/buildbot/#/builders/195/builds/13641 https://lab.llvm.org/buildbot/#/builders/197/builds/8319

Will put a fix up later today

Michael137 · 2025-08-24T09:26:11Z

Will need to go through a proper round of reviews but here's the preliminary patch: #155153

Starting with llvm#154686 the compressed_pair children are now wrapped in an anonymous structure. This patch adjusts the LLDB data-formatters to support that.

philnik777 · 2025-08-25T07:37:51Z

/cherry-pick f5e687d

llvmbot · 2025-08-25T07:43:48Z

Failed to cherry-pick: f5e687d

https://github.com/llvm/llvm-project/actions/runs/17202429330

Please manually backport the fix and push it to your github fork. Once this is done, please create a pull request

philnik777 · 2025-08-25T07:53:45Z

@tru This should cherry-pick cleanly once #154945 is merged. Do you have any preference on how exactly this should be done? Just wait for the other PR to be merged and then create a new PR, or cherry-pick both onto a branch and let them be merged together? Something different?

…PAIR (llvm#154686) LLVM 20 contained an ABI break that can result in the size of `std::unordered_{map,set,multimap,multiset}` and `std::deque` changing when used with an allocator type that is empty and contains a base class that is the same across rebound allocator instantiations (e.g. ``Allocator<int>`` and ``Allocator<char>`` are both empty and contain the same base class). In addition, the layout of a user-defined type that: - contains one of the following containers: `std::unordered_{map,set,multimap,multiset}`, `std::deque`, `std::map`, `std::set`, `std::multimap`, `std::multiset`, `std::list` or `std::vector`, and - passes an empty allocator, comparator or hasher type to that container, and - has a member of that same empty allocator, comparator or hasher type inside the enclosing struct, and - that member is either marked with `[[no_unique_address]]` or optimized out via the EBO (empty base optimization) technique saw its size increase from LLVM 19 to LLVM 20. This was caused by the usage of `[[no_unique_address]]` within some of libc++'s containers in a way that allowed subtle interactions with enclosing objects. This is fixed in LLVM 21 on Clang (returning to the LLVM 19 ABI), however that implies an ABI break from LLVM 20 to LLVM 21. Furthermore, fixing this causes a slight regression to constant evaluation support in `std::unique_ptr`. Specifically, constant evaluation will now fail when the deleter relies on being value-initialized for constant-evaluation admissibility. If a default-initialized deleter can be used during constant evaluation, or if the default constructor is non-trivial, the `unique_ptr` is not affected by this regression. In particular, this regression does not impact any `unique_ptr` using the default deleter. Note that there is currently no way to realistically fix this ABI break on GCC, therefore GCC will remain on the ABI introduced in LLVM 19. That also means that Clang and GCC will have a slightly different ABI for the small subset of types listed above until we are able to apply the same fix we did with Clang on GCC. We fix this regression by surrounding the members of the `_LIBCPP_COMPRESSED_PAIR` with an anonymous struct. This restricts the shifting of empty types to the front of the `_LIBCPP_COMPRESSED_PAIR` instead of throughout the surrounding object. This "frees up" the zero offset to contain another object of the same type, restoring the ability to perform EBO or to elide the storage for a type with `[[no_unique_address]]` in the enclosing (user-defined) struct. Fixes llvm#154146 Co-authored-by: Louis Dionne <[email protected]>

…155153) Starting with #154686 the compressed_pair children are now wrapped in an anonymous structure. This patch adjusts the LLDB data-formatters to support that. Outstanding questions: 1. Should GetChildMemberWithName look through anonymous structures? That will break users most likely. But maybe introducing a new API is worth it? Then we wouldnt have to do this awkward passing around of `anon_struct_index` 2. Do we support the layout without the anonymous structure? It's not too much added complexity. And we did release that version of libc++, so there is code out there compiled against it. But there is no great way of testing it (some of our macOS matrix bots do test it i suppose, but not in a targeted way). We have the layout "simulator" tests for some of the STL types which I will adjust.

…IR layout (#155153) Starting with llvm/llvm-project#154686 the compressed_pair children are now wrapped in an anonymous structure. This patch adjusts the LLDB data-formatters to support that. Outstanding questions: 1. Should GetChildMemberWithName look through anonymous structures? That will break users most likely. But maybe introducing a new API is worth it? Then we wouldnt have to do this awkward passing around of `anon_struct_index` 2. Do we support the layout without the anonymous structure? It's not too much added complexity. And we did release that version of libc++, so there is code out there compiled against it. But there is no great way of testing it (some of our macOS matrix bots do test it i suppose, but not in a targeted way). We have the layout "simulator" tests for some of the STL types which I will adjust.

h-vetinari · 2025-08-25T23:27:50Z

/cherry-pick f5e687d

llvmbot · 2025-08-25T23:34:22Z

/pull-request #155324

…lvm#155153) Starting with llvm#154686 the compressed_pair children are now wrapped in an anonymous structure. This patch adjusts the LLDB data-formatters to support that. Outstanding questions: 1. Should GetChildMemberWithName look through anonymous structures? That will break users most likely. But maybe introducing a new API is worth it? Then we wouldnt have to do this awkward passing around of `anon_struct_index` 2. Do we support the layout without the anonymous structure? It's not too much added complexity. And we did release that version of libc++, so there is code out there compiled against it. But there is no great way of testing it (some of our macOS matrix bots do test it i suppose, but not in a targeted way). We have the layout "simulator" tests for some of the STL types which I will adjust. (cherry picked from commit 20dd053)

…lvm#155153) Starting with llvm#154686 the compressed_pair children are now wrapped in an anonymous structure. This patch adjusts the LLDB data-formatters to support that. Outstanding questions: 1. Should GetChildMemberWithName look through anonymous structures? That will break users most likely. But maybe introducing a new API is worth it? Then we wouldnt have to do this awkward passing around of `anon_struct_index` 2. Do we support the layout without the anonymous structure? It's not too much added complexity. And we did release that version of libc++, so there is code out there compiled against it. But there is no great way of testing it (some of our macOS matrix bots do test it i suppose, but not in a targeted way). We have the layout "simulator" tests for some of the STL types which I will adjust. (cherry picked from commit 20dd053) (cherry picked from commit f3ec20b)

…IR layout (#155153) Starting with llvm/llvm-project#154686 the compressed_pair children are now wrapped in an anonymous structure. This patch adjusts the LLDB data-formatters to support that. Outstanding questions: 1. Should GetChildMemberWithName look through anonymous structures? That will break users most likely. But maybe introducing a new API is worth it? Then we wouldnt have to do this awkward passing around of `anon_struct_index` 2. Do we support the layout without the anonymous structure? It's not too much added complexity. And we did release that version of libc++, so there is code out there compiled against it. But there is no great way of testing it (some of our macOS matrix bots do test it i suppose, but not in a targeted way). We have the layout "simulator" tests for some of the STL types which I will adjust. (cherry picked from commit 20dd053)

philnik777 changed the title ~~fix abi break~~ [libc++] Fix ABI break introduced by switching to _LIBCPP_COMPRESSED_PAIR Aug 21, 2025

philnik777 force-pushed the fix_abi_break branch 2 times, most recently from 6f3c374 to 5d9b6ae Compare August 21, 2025 11:54

ldionne reviewed Aug 21, 2025

View reviewed changes

philnik777 force-pushed the fix_abi_break branch 3 times, most recently from 92355d0 to b07285a Compare August 22, 2025 13:27

ldionne marked this pull request as ready for review August 22, 2025 13:27

ldionne requested a review from a team as a code owner August 22, 2025 13:27

llvmbot added the libc++ libc++ C++ Standard Library. Not GNU libstdc++. Not libc++abi. label Aug 22, 2025

ldionne reviewed Aug 22, 2025

View reviewed changes

libcxx/docs/ReleaseNotes/21.rst Outdated Show resolved Hide resolved

ldionne reviewed Aug 22, 2025

View reviewed changes

ldionne added this to the LLVM 21.x Release milestone Aug 22, 2025

github-project-automation bot added this to LLVM Release Status Aug 22, 2025

github-project-automation bot moved this to Needs Triage in LLVM Release Status Aug 22, 2025

philnik777 force-pushed the fix_abi_break branch from ece477e to 838f78a Compare August 22, 2025 15:06

philnik777 commented Aug 22, 2025

View reviewed changes

libcxx/test/libcxx/containers/associative/map/abi.compile.pass.cpp Outdated Show resolved Hide resolved

philnik777 force-pushed the fix_abi_break branch 2 times, most recently from 275999a to 67fc781 Compare August 22, 2025 15:55

ldionne approved these changes Aug 22, 2025

View reviewed changes

github-project-automation bot moved this from Needs Triage to Needs Merge in LLVM Release Status Aug 22, 2025

ldionne mentioned this pull request Aug 22, 2025

[libc++] Vendor communication: upcoming ABI break in various container types #154985

Open

[libc++] Fix ABI break introduced by switching to _LIBCPP_COMPRESSED_…

0c5029c

…PAIR Update libcxx/docs/ReleaseNotes/21.rst Co-authored-by: Louis Dionne <[email protected]>

philnik777 force-pushed the fix_abi_break branch from 67fc781 to 0c5029c Compare August 22, 2025 18:32

philnik777 merged commit f5e687d into llvm:main Aug 23, 2025
72 of 79 checks passed

github-project-automation bot moved this from Needs Merge to Done in LLVM Release Status Aug 23, 2025

h-vetinari reviewed Aug 23, 2025

View reviewed changes

slydiman mentioned this pull request Aug 24, 2025

[TableGen][DecoderEmitter] Repurpose Filter class #155065

Merged

Michael137 mentioned this pull request Aug 24, 2025

[lldb][DataFormatters] Support newer _LIBCPP_COMPRESSED_PAIR layout #155153

Merged

llvmbot added the release:cherry-pick-failed label Aug 25, 2025

llvmbot removed the release:cherry-pick-failed label Aug 25, 2025

h-vetinari mentioned this pull request Aug 26, 2025

release/21.x: [libc++] Fix ABI break introduced by switching to _LIBCPP_COMPRESSED_PAIR (#154686) #155324

Closed

Michael137 mentioned this pull request Aug 26, 2025

🍒[lldb][DataFormatters] Support newer _LIBCPP_COMPRESSED_PAIR layout swiftlang/llvm-project#11257

Closed

Michael137 mentioned this pull request Aug 26, 2025

🍒[lldb][DataFormatters] Support newer _LIBCPP_COMPRESSED_PAIR layout swiftlang/llvm-project#11258

Closed

Michael137 mentioned this pull request Aug 26, 2025

🍒[lldb][DataFormatters] Support newer _LIBCPP_COMPRESSED_PAIR layout swiftlang/llvm-project#11259

Merged

dwblaikie mentioned this pull request Sep 16, 2025

port 5b4819e to release #159209

Merged

jhuber6 mentioned this pull request Sep 17, 2025

[LLVM] Update CUDA ELF flags for their new ABI #159451

Merged

[libc++] Fix ABI break introduced by switching to _LIBCPP_COMPRESSED_PAIR #154686

[libc++] Fix ABI break introduced by switching to _LIBCPP_COMPRESSED_PAIR #154686

Uh oh!

Conversation

philnik777 commented Aug 21, 2025 • edited by ldionne Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Aug 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ldionne Aug 21, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

llvmbot commented Aug 22, 2025

Uh oh!

Uh oh!

ldionne Aug 22, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ldionne left a comment

Choose a reason for hiding this comment

Uh oh!

philnik777 commented Aug 22, 2025

Uh oh!

Uh oh!

h-vetinari Aug 23, 2025

Choose a reason for hiding this comment

Uh oh!

philnik777 Aug 23, 2025

Choose a reason for hiding this comment

Uh oh!

h-vetinari Aug 23, 2025

Choose a reason for hiding this comment

Uh oh!

boomanaiden154 commented Aug 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Michael137 commented Aug 23, 2025

Uh oh!

slydiman commented Aug 24, 2025

Uh oh!

Michael137 commented Aug 24, 2025

Uh oh!

Michael137 commented Aug 24, 2025

Uh oh!

philnik777 commented Aug 25, 2025

Uh oh!

llvmbot commented Aug 25, 2025

Uh oh!

philnik777 commented Aug 25, 2025

Uh oh!

h-vetinari commented Aug 25, 2025

Uh oh!

llvmbot commented Aug 25, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

philnik777 commented Aug 21, 2025 •

edited by ldionne

Loading

github-actions bot commented Aug 21, 2025 •

edited

Loading

boomanaiden154 commented Aug 23, 2025 •

edited

Loading