On Fri, May 25, 2018 at 2:49 PM, Raymond Burkholder via Boost-users < boost-users@lists.boost.org> wrote:
Would it be possible for you to post a modified snippet of your code showing your new way of doing things?
I have changed many things and that wasn't real code. But the easiest way to transform my example would be to change the header size. I went from having a CRLF limited line to a 64 bit integer, which I serialize right into the buffer. This way I always know how large it is and don't need read_until anymore. Like this: void read_header_size() { asio::async_read(m_socket, m_streambuf, asio::transfer_at_least(8), [self](const boost::system::error_code, const std::size_t) { // < handle error skipped > boost::uint64_t header_size = 0; std::istream is(&self->m_streambuf, std::ios::binary); is.read(reinterpret_cast<char *>(&header_size), sizeof(header_size)); self->read_header(header_size); }); } The rest could stay as it is. We're not mixing the functions anymore. I changed more though. Gave my packet a method to de-serialize itself from a streambuf and return the number of missing bytes or 0 if ready. Then I call this until it returns a number of bytes and only then go for more async read, with the handler calling the de-serialize again. Advantages are less code and I can use the same method on client and server alike. Cheers, Stephan