c++20/libstdc++: `char_traits.h:213:13: note: subexpression not valid in a constant expression`

[timblechmann]

using this constexpr boost static_string (requires static_string develop) does not compile with clang against libstdc++ (g++ compiles against libstdc++ also clang++ compiles against libc++).

#include <boost/static_string.hpp>
using str = boost::static_strings::static_string<38>;
constexpr str a{ "mystr" };

c.cpp:5:15: error: constexpr variable 'a' must be initialized by a constant expression
constexpr str a{ "mystr" };
              ^~~~~~~~~~~~
/usr/bin/../lib/gcc/x86_64-linux-gnu/12/../../../../include/c++/12/bits/char_traits.h:213:13: note: subexpression not valid in a constant expression
   if (__s1 == __s2)
            ^
/usr/bin/../lib/gcc/x86_64-linux-gnu/12/../../../../include/c++/12/bits/char_traits.h:417:11: note: in call to 'move(&a.static_string_base::data_[0], &"mystr"[0], 5)'
   return __gnu_cxx::char_traits<char_type>::move(__s1, __s2, __n);
          ^
/home/tim/NIBuild/3rdparty/boost-v1.78.0-R8/boost/static_string/static_string.hpp:6000:3: note: in call to 'move(&a.static_string_base::data_[0], &"mystr"[0], 5)'
  traits_type::move(data(), s, size());
  ^
/home/tim/NIBuild/3rdparty/boost-v1.78.0-R8/boost/static_string/static_string.hpp:1413:12: note: in call to '&a->assign(&"mystr"[0], 5)'
    return assign(s, traits_type::length(s));
           ^
/home/tim/NIBuild/3rdparty/boost-v1.78.0-R8/boost/static_string/static_string.hpp:961:5: note: in call to '&a->assign(&"mystr"[0])'
    assign(s);
    ^
c.cpp:5:15: note: in call to 'basic_static_string(&"mystr"[0])'
constexpr str a{ "mystr" };
              ^
1 error generated.

the offending code in libstdc++ is this one:

	  if (__s1 == __s2) // unlikely, but saves a lot of work
	    return __s1;

https://github.com/gcc-mirror/gcc/blob/master/libstdc%2B%2B-v3/include/bits/char_traits.h#L216-L217

[llvmbot]

@llvm/issue-subscribers-c-20

[llvmbot]

@llvm/issue-subscribers-clang-frontend

[tbaederr]

Also reproducible using:

constexpr const char *f (const char* a, const char *b) {
  if (a == b)
    return a;

  return b;
}


constexpr const char *k = f("a", "b");

[tbaederr]

And we return failure (and print the diagnostic) because:

llvm-project/clang/lib/AST/ExprConstant.cpp

Lines 12955 to 12960 in be369ea

	// It's implementation-defined whether distinct literals will have
	// distinct addresses. In clang, the result of such a comparison is
	// unspecified, so it is not a constant expression. However, we do know
	// that the address of a literal will be non-null.
	if ((IsLiteralLValue(LHSValue) || IsLiteralLValue(RHSValue)) &&
	LHSValue.Base && RHSValue.Base)

This was implemented in 0683c0e, so let's ping the reviewers:
@zygoloid, @AaronBallman, @majnemer, @rnk, @compnerd, @rjmccall

[AaronBallman]

Heh, that's a fun one. http://eel.is/c++draft/lex.string#9 is what that code is referring to and I believe Clang is correct to reject it as a core constant expression per http://eel.is/c++draft/expr.const#5.24 because the equality operator's results are unspecified.

[jwakely]

This is a libstdc++ bug.

[jwakely]

And a g++ bug, for letting me do it.

[AaronBallman]

Thank you for confirming @jwakely -- I'm closing this as not a bug.