[libcxx] Remove an unused overload in filesystem_common.h. NFC.

[mstorsjo]

Summary

D96986 refactors this bit entirely though.

[curdeius]

LGTM.

[• Quuxplusone]

Sure, assuming it passes tests.
However, eyeballing the uses of this thing, I got to directory_iterator.cpp, which does

err.report(m_ec, "at root \"%s\"", root);

That produces garbage on Windows, right? because root is a path and the way we unwrap(path const&) is to turn it into a wchar_t*, not a char*?

[mstorsjo]

In D98065#2607603, @Quuxplusone wrote:

Sure, assuming it passes tests.
However, eyeballing the uses of this thing, I got to directory_iterator.cpp, which does

err.report(m_ec, "at root \"%s\"", root);

That produces garbage on Windows, right? because root is a path and the way we unwrap(path const&) is to turn it into a wchar_t*, not a char*?

Oh, you're right - I had missed that one (I have the same issue fixed in a couple other cases in operations.cpp). Thanks! I tried applying D96986 and then I do get format warnings that point out exactly this.

I'll get a fix posted for that issue.

[• Quuxplusone]

This is superseded by D98097, right?

[mstorsjo]

In D98065#2624007, @Quuxplusone wrote:

This is superseded by D98097, right?

Yes - I guess I can close this one right away.

libcxx/src/filesystem/filesystem_common.h

static string format_string_imp(const char* msg, ...) {
   size_with_null = static_cast<size_t>(ret) + 1;
   result.__resize_default_init(size_with_null - 1);
   ret = ::vsnprintf(&result[0], size_with_null, msg, args);
   _LIBCPP_ASSERT(static_cast<size_t>(ret) == (size_with_null - 1), "TODO");
 
   return result;
 }
 
-const path::value_type* unwrap(path::string_type const& s) { return s.c_str(); }
 const path::value_type* unwrap(path const& p) { return p.native().c_str(); }
 template <class Arg>
 Arg const& unwrap(Arg const& a) {
   static_assert(!is_class<Arg>::value, "cannot pass class here");
   return a;
 }
 
 template <class... Args>