[libc++abi] Use __has_feature check to enable usage of thread_local for exception storage #97591
Conversation
…or exception storage Previously, we'd use HAS_THREAD_LOCAL which was never defined. Hence, we'd basically never use the code path where we use thread_local. Fixes llvm#78207
|
@llvm/pr-subscribers-libcxxabi Author: Louis Dionne (ldionne) ChangesPreviously, we'd use HAS_THREAD_LOCAL which was never defined. Hence, we'd basically never use the code path where we use thread_local. Fixes #78207 Full diff: https://github.com/llvm/llvm-project/pull/97591.diff 1 Files Affected:
diff --git a/libcxxabi/src/cxa_exception_storage.cpp b/libcxxabi/src/cxa_exception_storage.cpp
index 2479f550e09ef..c842da195accb 100644
--- a/libcxxabi/src/cxa_exception_storage.cpp
+++ b/libcxxabi/src/cxa_exception_storage.cpp
@@ -24,7 +24,7 @@ extern "C" {
} // extern "C"
} // namespace __cxxabiv1
-#elif defined(HAS_THREAD_LOCAL)
+#elif __has_feature(cxx_thread_local)
namespace __cxxabiv1 {
namespace {
|
Thanks; I tested this patch in some of my test environments, and it seems to not break anything, so I think it's fine. As mentioned in #79542 (comment), this could have an effect on emutls environments; I tried to run a set of tests with emutls enabled as well. With emutls enabled (as in that linked PR), a bunch of my tests hang, both with and without this PR. So that doesn't indicate any reason to hold back this patch either. So, in short, +1 for this patch from me, and AFAIK this will fix the same issue as https://reviews.llvm.org/D155278 tried to fix. |
|
In case anyone else wonders why their binaries that statically link libc++abi no longer (No action needed for libc++abi, just mentioning it in case someone else runs into this.) |
…or exception storage (llvm#97591) Previously, we'd use HAS_THREAD_LOCAL which was never defined. Hence, we'd basically never use the code path where we use thread_local. Fixes llvm#78207
Previously, we'd use HAS_THREAD_LOCAL which was never defined. Hence, we'd basically never use the code path where we use thread_local.
Fixes #78207