From: Bertram Scharpf <lists@...>
Date: 2005-01-07T07:43:28+09:00
Subject: Re: HTTP doesn't work. A bug?

Hi,

Am Donnerstag, 30. Dez 2004, 08:43:52 +0900 schrieb Bertram Scharpf:
> Hi all,
> 
> downloading a larger file, I encounter this error message:
> 
> /usr/lib/ruby/1.8/net/protocol.rb:197:in `sysread': End of file reached (EOFError)
> 
> The error happens only with a download through a modem. In
> my small home network everything works properly. I use
> Apache 1.3.26. Outside, I tried to load from 1.3.29 and 2.0
> versions.

I examined the problem thoroughly and now, I can reproduce
it reliably.

What I have to do is starting several instances of

    wget -o /dev/null -O /dev/null http://www.ruby-doc.org/downloads/ruby-doc-bundle.tar.gz

Then, the program below dies when loading the first file the
second time.

I tried it with a maximum of debugging output, I even hacked
the interpreter to get some more. I diff'ed the results
against those of the run without the parallel downloads.
I didn't find any hint.

Of course, the program works when I omit the `conn.start'
statement, but I think it should get along with it.

The problem seems to be compiler-independent (GNU C 2.95
and 3.3).

Could anyone who knows more about socket connections please
have a look at the problem. I will provide any further
information.

Bertram


------------------------------

require 'net/http'

s, t = [ 'www.rayninfo.co.uk', 'www.ruby-doc.org']
f, g = [ '/vimtips.html',      '/docs/CLR-FAQ/ruby_newsgroup_faq.txt']

c, d = [ s, t].collect { |host| Net::HTTP.new host }

[ c, d].each { |conn| conn.start }

100.times do |n|
  [ [c,f], [d,g]].each do |conn,file|
    conn.request_get file do |resp|
      resp.read_body do |buf|
      end
    end
  end
end


-- 
Bertram Scharpf
Stuttgart, Deutschland/Germany
http://www.bertram-scharpf.de