[libc] Fix definition of UINT_MAX in limits.h

[jhuber6]

Summary:
Currently we use (~0U) for this definition, however the ~ operator
returns a different sign, meaning that preprocessor checks against this
value will fail. See https://godbolt.org/z/TrjaY1d8q where the
preprocessor thinks that it's not 0xffffffff while the static
assertion thinks it is. This is because the latter does implicit
conversion but the preprocessor does not. This is now consistent with
other headers.

[llvmbot]

@llvm/pr-subscribers-libc

Author: Joseph Huber (jhuber6)

Changes

Summary:
Currently we use (~0U) for this definition, however the ~ operator
returns a different sign, meaning that preprocessor checks against this
value will fail. See https://godbolt.org/z/TrjaY1d8q where the
preprocessor thinks that it's not 0xffffffff while the static
assertion thinks it is. This is because the latter does implicit
conversion but the preprocessor does not. This is now consistent with
other headers.

---

Full diff: https://github.com/llvm/llvm-project/pull/95279.diff

1 Files Affected:

• (modified) libc/include/llvm-libc-macros/limits-macros.h (+3-3)

diff --git a/libc/include/llvm-libc-macros/limits-macros.h b/libc/include/llvm-libc-macros/limits-macros.h
index 95f0f5f0baa58..3fab996b61ac9 100644
--- a/libc/include/llvm-libc-macros/limits-macros.h
+++ b/libc/include/llvm-libc-macros/limits-macros.h
@@ -148,7 +148,7 @@
 #endif // INT_MAX
 
 #ifndef UINT_MAX
-#define UINT_MAX (~0U)
+#define UINT_MAX (INT_MAX * 2U + 1U)
 #endif // UINT_MAX
 
 #ifndef LONG_MAX
@@ -160,7 +160,7 @@
 #endif // LONG_MAX
 
 #ifndef ULONG_MAX
-#define ULONG_MAX (~0UL)
+#define ULONG_MAX (LONG_MAX * 2UL + 1UL)
 #endif // ULONG_MAX
 
 #ifndef LLONG_MAX
@@ -172,7 +172,7 @@
 #endif // LLONG_MAX
 
 #ifndef ULLONG_MAX
-#define ULLONG_MAX (~0ULL)
+#define ULLONG_MAX (LLONG_MAX * 2ULL + 1ULL)
 #endif // ULLONG_MAX
 
 // *_MIN macros

[lntue]

shall we explicitly cast ((unsigned) INT_MAX)?

[jhuber6]

Should be handled by the 2U part, otherwise it would fail in the myriad templates we use in limits.h I believe. This is how Clang's implementation does it at least.

[MaskRay]

What's the issue?

#define UINT_MAX (~0U)
_Static_assert(UINT_MAX == 0xFFFFFFFF, "");

compiles.

[jhuber6]

What's the issue?

#define UINT_MAX (~0U)
_Static_assert(UINT_MAX == 0xFFFFFFFF, "");

compiles.

It's explained in the summary, ~0U is not seem as equal to 0xFFFFFFFF by the preprocessor in https://godbolt.org/z/TrjaY1d8q. This causes the situation shown there where the preprocessor says it's not 0xFFFFFFFF but the static assert says it is. I found this while compiling the compiler-rt which checks exactly this, furthermore this is how both GNU and Clang's resource headers seem to do it.

[MaskRay]

I see. Perhaps just #define UINT_MAX 0xffffffffU

[jhuber6]

I see. Perhaps just #define UINT_MAX 0xffffffffU

Some targets have UINT_MAX as something other than 32-bits, The compiler defines __INT_MAX__ which tends to tell us the base width of an int, and thus an unsigned is just twice that plus one.