Optimize llformat Even MORE, and clean up some code.

Thanks for the help, Aru!

indra/llcommon/llformat.cpp

@@ -36,21 +36,35 @@
 #include "llformat.h"
 
 #include <cstdarg>
+#include <boost/align/aligned_allocator.hpp>
 
 // common used function with va_list argument
 // wrapper for vsnprintf to be called from llformatXXX functions.
-static void va_format(std::string& out, const char *fmt, va_list va)
+static void va_format(std::string& out, const char *fmt, va_list& va)
 {
-	constexpr auto smallsize = 1024;
-	std::vector<char> charvector(smallsize); // Evolves into charveleon
+	typedef typename std::vector<char, boost::alignment::aligned_allocator<char, 1>> vec_t;
+	static thread_local vec_t charvector(1024); // Evolves into charveleon
+	#define vsnprintf(va) std::vsnprintf(charvector.data(), charvector.capacity(), fmt, va)
+#ifdef LL_WINDOWS // We don't have to copy on windows
+	#define va2 va
+#else
 	va_list va2;
 	va_copy(va2, va);
-	const auto size = std::vsnprintf(charvector.data(), charvector.size(), fmt, va);
-	if (size >= smallsize)
+#endif
+	const auto smallsize(charvector.capacity());
+	const auto size = vsnprintf(va);
+	if (size < 0)
+	{
+		LL_ERRS() << "Encoding failed, code " << size << ". String hint:" << out << '/' << fmt << LL_ENDL;
+	}
+	else if (static_cast<vec_t::size_type>(size) >= smallsize) // Resize if we need more space
 	{
 		charvector.resize(1+size); // Use the String Stone
-		std::vsnprintf(charvector.data(), charvector.size(), fmt, va2);
+		vsnprintf(va2);
 	}
+#ifndef LL_WINDOWS
+	va_end(va2);
+#endif
 	out.assign(charvector.data());
 }