encoding/binary's API can be a bit cumbersome and its performance leaves quite a bit to be desired. This is a re-imagining of that library.
Use of io.Reader and io.Writer seems like an unfortunate decision in encoding/binary. A common pattern when using the library is allocating a buffer for the encoded data and wrapping it in a bytes.Buffer in order to pass it to binary.Read or binary.Write. Internally, these functions allocate another buffer for the full encoding and copy between it and the io.Reader or io.Writer.
Another irritating consequence of using io.Reader and io.Writer is that it proxies your errors unnecessarily. This was actually the original motivation of this library. This library was originally written as part of the gVisor project which uses its own error type.
In the included benchmark, this use of this library imparts roughly a 5x speedup over encoding/binary. This is almost entirely due to giving the library a more reasonable API, further optimization is almost certainly possible.
Please note that this library panics if the buffer provided to Unmarshal is not exactly the correct size. The reasoning for this is that passing an incorrect size is programmer error and not a dynamic runtime condition. Marshal does not have this property.