Looks like the issue has already been reported upstream: https://sourceware.org/bugzilla/show_bug.cgi?id=31320

From personal testing, the following path was enough to get it to build on my personal machine (Sonoma, arm64):

diff --git a/gold/output.cc b/gold/output.cc
index ead67f20363..98b2f25c18b 100644
--- a/gold/output.cc
+++ b/gold/output.cc
@@ -29,7 +29,9 @@
 #include <unistd.h>
 #include <sys/stat.h>
 #include <algorithm>
+#ifndef __APPLE__
 #include <uchar.h>
+#endif
 
 #ifdef HAVE_SYS_MMAN_H
 #include <sys/mman.h>
diff --git a/gold/stringpool.cc b/gold/stringpool.cc
index d8f38cfabc1..e04909edad1 100644
--- a/gold/stringpool.cc
+++ b/gold/stringpool.cc
@@ -25,7 +25,9 @@
 #include <cstring>
 #include <algorithm>
 #include <vector>
+#ifndef __APPLE__
 #include <uchar.h>
+#endif
 
 #include "output.h"
 #include "parameters.h"

Using this suggestion from the thread did not work for me:

#ifdef __APPLE__
#include <stdint.h>
typedef uint_least16_t char16_t;
typedef uint_least32_t char32_t;
#else
#include <uchar.h>
#endif

output.cc:34:24: error: cannot combine with previous 'type-name' declaration specifier
typedef uint_least16_t char16_t;
                       ^
output.cc:35:24: error: cannot combine with previous 'type-name' declaration specifier
typedef uint_least32_t char32_t;
                       ^

Rerunning to see which SDKs have the header and which don't.

Testing the patch at https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=941d02eaae0557c80c9e4130478e584a8b284494 mentioned in https://sourceware.org/bugzilla/show_bug.cgi?id=30867

I've made some tests in #165082 and I'm not sure there is a way for macOS 12. Newer binutils require C++11 with uchar.h available, and macOS 12 doesn't seem to have that.

Upstream effectively decided to raise their requirements and excluded Monterey and earlier. Not sure why it only manifests for some targets.

🤖 An automated task has requested bottles to be published to this PR.

Whoops I approved to fast, I wanted to discuss the Monterey situation first, sorry. I canceled the upload job.

Would this break some reverse deps on Monterey? I see a few formula that would still need it on macOS

Neither arm-linux-gnueabihf-binutils nor x86_64-linux-gnu-binutils are dependencies of anything in homebrew-core.

Unlike C, char16_t and char32_t are builtin types in C++. They are not defined in uchar.h so the include seems unnecessary entirely unless they're catering for C99-compatible compilers that are not C++11 compliant (though https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=941d02eaae0557c80c9e4130478e584a8b284494 seems to suggest otherwise).

For example, see:

• https://godbolt.org/z/Pc79Ksheb (oldest GCC on Godbolt which accepts -std=c++0x)
• https://godbolt.org/z/vb1cMb9x7 (Clang 3)
• https://godbolt.org/z/5fEb3dGbb (MSVC)
• https://godbolt.org/z/fTqr86W8T (icc)

It would be needed on GCC 4.1 which supports C99 but not C++11, but since GCC 4.1 doesn't accept -std=c++0x anyway such compiler would be blocked early: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=941d02eaae0557c80c9e4130478e584a8b284494.

So the correct patch for me is to remove those includes, given they are not compatible with libc++ prior to 2022 (15.0): llvm/llvm-project@311ff39 and https://sourceware.org/bugzilla/show_bug.cgi?id=30867 shows they are willing to support GCC from 2013.

I imagine this issue also affects some of the BSDs too.

As stated above in response to:

Would this break some reverse deps on Monterey? I see a few formula that would still need it on macOS

Neither arm-linux-gnueabihf-binutils nor x86_64-linux-gnu-binutils are dependencies of anything in homebrew-core.

Maybe upstream will accept a patch to improve the situation, but right now they effectively decided to require it, so I think we should just follow that.

🤖 An automated task has requested bottles to be published to this PR.

⚠️ Bottle publish failed.

@p-linnane has requested bottles to be published to this PR.