[libc] Workaround for GCC on typedef for _Complex __float128
#157010
Conversation
Currently, GCC can't parse `typedef _Complex __float128 cfloat128;`, although `__typeof__` can be used as a workaround. Reported https://gcc.gnu.org/PR121799. Some recent changes exposed it to GCC and then caused CI failure for libc++. This patch adds a workaround for GCC.
llvmbot
added
the
libc++
|
@llvm/pr-subscribers-libcxx @llvm/pr-subscribers-libc Author: A. Jiang (frederick-vs-ja) ChangesCurrently, GCC can't parse Some recent changes exposed it to GCC and then caused CI failure for libc++. This patch adds a workaround for GCC. Full diff: https://github.com/llvm/llvm-project/pull/157010.diff 2 Files Affected:
diff --git a/libc/include/llvm-libc-types/cfloat128.h b/libc/include/llvm-libc-types/cfloat128.h
index 25b4cc7345a30..78a3dacd2676c 100644
--- a/libc/include/llvm-libc-types/cfloat128.h
+++ b/libc/include/llvm-libc-types/cfloat128.h
@@ -13,7 +13,12 @@
#ifdef LIBC_TYPES_HAS_CFLOAT128
#ifndef LIBC_TYPES_CFLOAT128_IS_COMPLEX_LONG_DOUBLE
+#if defined(__GNUC__) && !defined(__clang__)
+// Remove the workaround when https://gcc.gnu.org/PR121799 gets fixed.
+typedef __typeof__(_Complex __float128) cfloat128;
+#else // ^^^ workaround / no workaround vvv
typedef _Complex __float128 cfloat128;
+#endif // ^^^ workaround ^^^
#else
typedef _Complex long double cfloat128;
#endif // LIBC_TYPES_CFLOAT128_IS_COMPLEX_LONG_DOUBLE
diff --git a/libcxx/include/__config b/libcxx/include/__config
index c197851f1c8fe..b1dff47b6aafd 100644
--- a/libcxx/include/__config
+++ b/libcxx/include/__config
@@ -1265,3 +1265,5 @@ typedef __char32_t char32_t;
#endif // __cplusplus
#endif // _LIBCPP___CONFIG
+
+// triggering CI, will be reverted
|
| #ifdef LIBC_TYPES_HAS_CFLOAT128 | ||
| #ifndef LIBC_TYPES_CFLOAT128_IS_COMPLEX_LONG_DOUBLE | ||
| #if defined(__GNUC__) && !defined(__clang__) | ||
| // Remove the workaround when https://gcc.gnu.org/PR121799 gets fixed. |
There was a problem hiding this comment.
This was considered to be duplicate of https://gcc.gnu.org/PR32187 later. Will update the comment once it's confirmed the CI failure due to libc is fixed.
| // Remove the workaround when https://gcc.gnu.org/PR121799 gets fixed. | |
| // Remove the workaround when https://gcc.gnu.org/PR32187 gets fixed. |
There was a problem hiding this comment.
Done. It can be confirmed that GCC is happy now. https://github.com/llvm/llvm-project/actions/runs/17483703784/job/49660001482
|
Just out of curiosity, does libc++ already incorporate llvm-libc? I recall seeing an earlier PR that tried to use llvm-libc in APFloat, but it hasn't been merged yet... |
Yes. Some components provided by libc++'s |
frederick-vs-ja
removed
the
libc++
3 participants
Currently, GCC can't parse
typedef _Complex __float128 cfloat128;, although__typeof__can be used as a workaround. Reported https://gcc.gnu.org/PR121799 which was later considered as duplicate of https://gcc.gnu.org/PR32187.Some recent changes exposed it to GCC and then caused CI failure for libc++. This patch adds a workaround for GCC.