[libc++][z/OS] Need to define _LIBCPP_HAS_UNICODE to 0 for EBCDIC

[zibi2]

This PR is needed since EBCDIC does not support UNICODE.

[llvmbot]

@llvm/pr-subscribers-libcxx

Author: Zibi Sarbinowski (zibi2)

Changes

This PR is needed since EBCDIC does not support UNICODE.

---

Full diff: https://github.com/llvm/llvm-project/pull/119244.diff

1 Files Affected:

• (modified) libcxx/include/__config_site.in (+4)

diff --git a/libcxx/include/__config_site.in b/libcxx/include/__config_site.in
index fc01aaf2d8746e..b575a5de8e0d0b 100644
--- a/libcxx/include/__config_site.in
+++ b/libcxx/include/__config_site.in
@@ -28,7 +28,11 @@
 #cmakedefine01 _LIBCPP_HAS_FILESYSTEM
 #cmakedefine01 _LIBCPP_HAS_RANDOM_DEVICE
 #cmakedefine01 _LIBCPP_HAS_LOCALIZATION
+#if defined(__MVS__) && !defined(__NATIVE_ASCII_F)
+#cmakedefine _LIBCPP_HAS_UNICODE 0
+#else
 #cmakedefine01 _LIBCPP_HAS_UNICODE
+#endif
 #cmakedefine01 _LIBCPP_HAS_WIDE_CHARACTERS
 #cmakedefine _LIBCPP_HAS_NO_STD_MODULES
 #cmakedefine01 _LIBCPP_HAS_TIME_ZONE_DATABASE

[perry-ca]

LGTM.

[ldionne]

Supporting EBCDIC and z/OS requires a lot more changes than just this unless I am mistaken. I know you folks tried upstreaming a lot of that support a few years ago but AFAICT we never got consensus for moving forward with those changes, which were quite involved. AIX doesn't have the same challenges and we're happy to support it upstream.

I would like to have a broader discussion about EBCDIC support in libc++ before we start cherry-picking small changes like this into our codebase -- that ask is consistent with what we do for other platforms that are not "officially supported".