updated readme

README.md

@@ -10,11 +10,7 @@ The easiest way is to use built-in CMake FetchContent:
 
 ```cmake
 include(FetchContent)
-FetchContent_Declare(
-    ecpp_endian
-    GIT_REPOSITORY https://github.com/jmacheta/endian.git
-    GIT_TAG main
-)
+FetchContent_Declare(ecpp_endian URL https://github.com/jmacheta/endian/tarball/latest)
 
 FetchContent_MakeAvailable(ecpp_endian)
 ```
@@ -26,5 +22,26 @@ target_link_libraries(my_target PUBLIC ecpp::endian)
 ```
 
 ## Usage
+### byteswap - unconditionally reserse byte order in integral types:
+```cpp
+#include <ecpp/endian.hpp>
 
-todo
+uint64_t a = 0x1122'3344'5566'7788U;
+static_assert(0x8877'6655'4433'2211U == ecpp::byteswap(a));
+
+```
+
+### hton / ntoh - convert endiannes from/to network endianness (big-endiann)
+```cpp
+// On little endian systems:
+uint64_t a = 0x1122'3344'5566'7788U;
+static_assert(0x8877'6655'4433'2211U == ecpp::hton(a));
+static_assert(a == ecpp::ntoh(ecpp::hton(a)));
+
+// On big endian systems:
+uint64_t a = 0x1122'3344'5566'7788U;
+static_assert(a == ecpp::hton(a));
+static_assert(a == ecpp::ntoh(ecpp::hton(a)));
+```
+## Limitations
+While every function is marked as `constexpr`, in C++ standards prior to C++20, this might be untrue due to the non-constexpr `std::memcpy`. I have checked many clang/gcc versions, and this does not seem to be an issue, but it's worth noting.

include/ecpp/endian.hpp

@@ -18,7 +18,7 @@ template <std::integral T> [[nodiscard]] constexpr T byteswap(T value) noexcept
   return std::bit_cast<T>(value_representation);
 }
 #else
-template <std::integral T> [[nodiscard]] inline T byteswap(T value) noexcept {
+template <std::integral T> [[nodiscard]] constexpr T byteswap(T value) noexcept {
   alignas(T) std::array<std::byte, sizeof(T)> value_representation;
   std::memcpy(value_representation.data(), &value, sizeof(value));
   std::ranges::reverse_copy(value_representation, reinterpret_cast<std::byte *>(&value));