[libcxx-commits] [libcxx] [libc++] Optimize num_put integral functions (PR #120859)

[Nico Weber via libcxx-commits]

nico wrote:

Did you end up adding a release notes entry?

I reduced the `int256` thing mentioned above to

```
#include <algorithm>
#include <sstream>
#include <utility>

#include <gtest/gtest.h>

struct MyType {};

std::ostream& operator<<(std::ostream& o, const MyType& b) {
  return o << 0;
}

TEST(Int256, OStream) {
  struct {
    std::ios_base::fmtflags flags;
    const char* rep;
  } cases[] = {
      // showbase does nothing on zero
      { std::ios::dec | std::ios::showbase, "0"},
      { std::ios::oct | std::ios::showbase, "0"},
      { std::ios::hex | std::ios::showbase, "0"},
  };
  for (size_t i = 0; i < ABSL_ARRAYSIZE(cases); ++i) {
    std::ostringstream os;
    os.flags(cases[i].flags);
    MyType type;
    os << type;
    EXPECT_EQ(cases[i].rep, os.str());
  }
}
```


That used to pass; now it outputs `0`, `00`, and `0x0` instead. (I.e. showbase on `0` used to not add a prefix and now it does.)

I'm guessing that's also intentional?

https://github.com/llvm/llvm-project/pull/120859