New sockets.cs patch submission

Hello list,

Please find attached another patch for Socket.cs with the following

  1. Added class SocketStream : System.IO.Stream which wraps the socket
    and provides some stream-like functionality which RubyIO can use. I’ve
    tested the following methods and they appear to work well now on
  • readline
  • readlines
  • gets
    puts also works, but I’m not as confident that it’s correct - see
    below under “Issues”. I’ve also included a test file I used to compare
    output between ruby.exe and rbx.exe (I tested against a default
    install of IE7 - which returns a 404 for the address requested and
    both executables appear to produce the same output except for
    line-ending issues)
  1. Fixed up the spacing/tabbing issues in the file as per Seo’s
    comments - everything is now 4 spaces and no tabs, and I also moved
    the opening curly braces up to match the style in the rest of the

I know this code isn’t perfect - but I think it is a step in the right
direction. I have some concerns about it which I’ve listed below - any
comments or suggestions will be most welcome.


  • Can only peek 1 byte - there is no buffering of data off the socket
  • I don’t know if this will have a performance / efficiency impact?
  • There is no error handling code around the socket calls - is this
    OK? If not, what should be done?
  • The third parameter passed to RubyIO’s constructor - “r+” - what
    should this be? “r+” allows me to read and write to the socket through
    the stream interface so maybe it’s OK?
  • None of the IO mode checking referred to in the TODO comment above
    the SocketStream class has been implemented - I presume this means
    checking whether the socket has been shutdown either for sending or
    receiving before actually trying to do the read/write - instead of
    letting it throw an exception.
  • How should I trigger a flush on write? At the moment it’s done on
    seeing a newline (hard-coded) in the stream, but I don’t think that’s
    clean or correct (certainly ruby.exe seems to send my test request
    (which has 2 newlines at the end) in one go - not in two separate
    pieces - which is the minor difference I referred to above regarding
  • Not sure if Flush is efficient - should I somehow use an array
    rather than a list of bytes?

Thanks for your time


By the way, the latest Rubinius specs have a host of specs for Socket et
Sadly most of them are empty at this stage. But at least it gives you
something to start testing with.