bug#12893: 24.2.50; url-retrieve fails for localhost

[Dmitry Gutov]

On 22.12.2012 17:52, Andreas Schwab wrote:
> You need to look at the url-proxy-services variable.

It has the default value of nil.

So, here's the small example of a web server in Ruby:

```
require 'webrick'

class Server < WEBrick::HTTPServer
  def initialize(port)
    super(Port: port)
    mount("/", Handler)
  end

  class Handler < WEBrick::HTTPServlet::AbstractServlet
    def do_GET(req, res)
      res["Content-Type"] = "text/html"
      res.status = 200
      res.body = "Hello!\n"
      raise WEBrick::HTTPStatus::OK
    end
  end
end

server = Server.new(54321)

%w(INT TERM).each { |s| trap(s) { server.stop } }

server.start
```

When I run it (ruby server.rb), I see the following output:

[2012-12-23 04:12:53] INFO  WEBrick 1.3.1
[2012-12-23 04:12:53] INFO  ruby 1.9.3 (2012-11-10) [x86_64-linux]
[2012-12-23 04:12:53] WARN  TCPServer Error: Address already in use - 
bind(2)
[2012-12-23 04:12:53] INFO  Server#start: pid=4838 port=54321

The third line probably hints at problem with the system, but

curl http://localhost:54321/

outputs the expected "Hello!", and running

lsof -i TCP:54321

doesn't show anything when the server isn't running.

Now, when the server is running, evaluating

(switch-to-buffer (url-retrieve-synchronously "http://localhost:54321/";))

shows an empty buffer, whereas

(switch-to-buffer (url-retrieve-synchronously "http://127.0.0.1:54321/";))

shows the expected contents.

Now, I couldn't reproduce the problem on my Windows machine, so the test 
case is probably not very useful.

If you tell me where to poke to solve the "address already in use" 
warning, I'll be sure to do that.