Bug #806: Net::Protocol#rbuf_fill is slow

• Status: Closed
• Assignee: -
• Target version: -

[tenderlovemaking (Aaron Patterson)]

=begin
Net::Protocol#rbuf_fill is slow. We can take advantage of io.read_nonblock and IO.select to deal with timeouts.

I've attached a patch.
=end

[matz (Yukihiro Matsumoto)]

=begin
merged. report if any problem caused by this patch.
=end

[matz (Yukihiro Matsumoto)]

• Status changed from Open to Closed
• % Done changed from 0 to 100

=begin
Applied in changeset r20443.
=end

[bitsweat (Jeremy Daer)]

=begin
Breaks if the IO doesn't respond to read_nonblock. StringIO, for example. Fall back to sysread in that case:

Index: lib/net/protocol.rb

--- lib/net/protocol.rb (revision 22024)
+++ lib/net/protocol.rb (working copy)
@@ -131,14 +131,20 @@
BUFSIZE = 1024 * 16

  def rbuf_fill

•  begin
  

•    @rbuf << @io.read_nonblock(BUFSIZE)
  

•  rescue Errno::EWOULDBLOCK
  

•    if IO.select([@io], nil, nil, @read_timeout)
  

•      retry
  

•    else
  

•      raise Timeout::TimeoutError
  

•  if @io.respond_to?(:read_nonblock)
  

•    begin
  

•      @rbuf << @io.read_nonblock(BUFSIZE)
  

•    rescue Errno::EWOULDBLOCK
  

•      if IO.select([@io], nil, nil, @read_timeout)
  

•        retry
  

•      else
  

•        raise Timeout::TimeoutError
  

•      end
     end
  

•  else
  

•    timeout(@read_timeout) do
  

•      @rbuf << @io.sysread(BUFSIZE)
  

•    end
   end
  

  end

=end

[matthewford (Matthew Ford)]

=begin
I can confirm that Jeremy Kemper's patch works for me and that the patch that was applied breaks IO that doesn't respond to read_nonblock.
=end