Network Transport Framework

Roadmap

The following outline describes the future direction of the project and general things to-do.

• Features
  • Implement something like "send all" and "receive all" functions for ntsi::StreamSocket to provide convenience to users of blocking sockets to send/receive all desired data possibly over multiple system calls
  • Support the specification of relative timeouts when describing deadlines for timers and socket operations
  • Support zero-copy transmissions on Linux
  • Support sending and receiving multiple datagram messages in a single system call, where supported (sengmmsg/recvmmsg)
  • Implement ntco::IoRing on Windows using their API
• Fixes
  • Detach a socket from a reactor asynchronously
  • Cancel all pending operations on a ntci::ProactorSocket asynchronously
• Optimization
  • Allow the user to mask the announcement of certain "events" detected passively during the operation of an asynchronous socket
  • Improve the implementation of ntci::Reactor using io_uring on Linux
  • Attempt to split the single mutex used by, e.g., ntcr::StreamSocket into two mutexes, one for reading and one for writing, so that readers don't block writers and vice-versa
  • Revise the virtual write queue data structures so that any potentially-batchable buffer array is built/adjusted at the time each send operation is initiated, rather than re-built from scratch each time a socket is detected to be writable
• Operations
  • Re-establish a contiguous integration process in an open-source context, possibly using GitHub Actions
  • Integrate with certain popular package managers (e.g. debian, rpm, conan, vcpkg, etc.)
  • Adopt the new BDE build system to replace the old BDE build system as an alternate method for build this project