On Thu, 12 Aug 2010, Luke Dashjr wrote:

>> As far as I remember, we've never had a non-contriversial and complete
>> patch for that problem. Can you provide the patch again please?
>
> Attached. One person had some compiler compatibility change suggestions,
> which was before the -r2 version of this patch. After that, one person wrote
> that it "works fine, passes all tests, but it is still plain wrong" but
> never came up with any real technical reasons *why* it was wrong.

Thanks. I can see a few problems with this patch:

- It still handles only one timeout stored at any given time for an easy
   handle, which feels like it risks dropping a timeout as we use 1 millisecond
   ones every now and then. I am the one who introduced this as a single
   timeout so I'm not blaming anyone else, I've just grown into believing it
   isn't good enough. Hence my approach to use a list for the timeouts and
   keep track of *all* timeouts set with Curl_expire().

- I don't see how this handles CURLOPT_CONNECTTIMEOUT so it isn't fixing the
   whole known bug #62.

- It seems to assume that the timeout struct field is always set to a sensible
   non-zero value, while the timeout value is in fact 0 (zero) when the timeout
   isn't used.

Since I believe I have come up with a better approach to this problem, I would
rather have us work on that and polish off any remaining problems it may
suffer from. I will merge those changes into mainline within a day or two and
then we can all just use git HEAD to test drive these modifications.

Thanks a lot for your help and interest to improve libcurl!

-- 
  / daniel.haxx.se
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html