From: Mathieu Bouchard <matju@...>
Date: 2001-12-03T12:48:06+09:00
Subject: [ruby-talk:27314] BUG in #read


With #read (at least TCPSocket, maybe IO), trying to read n bytes from a
non-blocking socket causes either n bytes to be returned or an exception
to be thrown. in the event that m bytes are read, 0<m, m<n, then m bytes
are *lost*.

This situation occurs when you have to use sockets on unknown quantities
of data. many protocols are designed so that you always know how much to
read (eg. X11); others no.

remedies:
1. #read should not lose data
2. #read or a version of #read (#read_at_most) should allow returning
incomplete results (result.length < n)

workaround (*SLOW*):
def bugfree_read_at_most n
  s=""
  (s << (read 1) while s.length<n) rescue Errno::EWOULDBLOCK
  s
end

________________________________________________________________
Mathieu Bouchard                   http://hostname.2y.net/~matju