[libc++][format] Should `operator<<(thread::id)` care about fmtflags in C++23?

[JMazurkiewicz]

Repro:

#include <cassert>
#include <format>
#include <sstream>
#include <thread>

using namespace std;

int main() {
  auto id = this_thread::get_id();
  const auto s1 = format("{}", id);
  ostringstream stream;
  stream << id;
  const auto s2 = stream.str(); // This is "text representation".
  assert(s2 == s1);             // Ok.

  stream.str("");
  stream << hex << id;
  const auto s3 =
      stream.str(); // Should this be the same "text representation"?
  assert(s3 == s1); // Currently, it is not.
}

Build:

PS D:\libcxx-playground> clang -std=c++2b -fexperimental-library -Iinclude\c++\v1 -llib\c++ "-llib\windows\clang_rt.builtins-x86_64" .\test.cpp
PS D:\libcxx-playground> .\a.exe
Assertion failed: s3 == s1, file .\test.cpp, line 20

The question is: should operator<<(thread::id) even care about fmtflags? It seems to me that since P2693R1 (partially implemented in 88622aa) the text representation is not longer "unspecified" and should always be the same, no matter what the fmtflags/locale/precision are.

CC: @mordante

[mordante]

This is an interesting question since in this example the output of operator<< is not identical either. I haven't checked, but I'm quite sure that was a requirement before this change. I will look into this.

[mordante]

Based on the wording prior to P2693R1 this requirement was true for operator<< too. I had a quick look at all major implementations and I expect them all the have the same issue. I will test further at another time.

[mordante]

https://reviews.llvm.org/D153336

[mordante]

This has been merged 1c1edd1