ok. I see.... thanks for explaining. I got excited as I did not know whole
history of changes.

I will think here and try to make a small sample program, but this will be
(as you correctly assumed) tricky (and that is why I need to think). as
minimum, I will describe use case in more detail (but your current
understanding of the use case is correct). If I am lucky, I will be able to
identify the actual problem in the code (still reconstructing state
transitions in multi_runsingle() for myself).

On Thu, Oct 9, 2014 at 3:09 PM, Daniel Stenberg <daniel_at_haxx.se> wrote:

> On Thu, 9 Oct 2014, Alibek Joraev wrote:
>
> I can confirm that this revision fixes this issue.
>> git checkout d242839af8511b389f0edd6519bdae6cd860e8a9
>>
>
> ...
>
> what is the next now? are you planning (and when) to release new patch or
>> version for Curl?
>>
>
> Hold your horses. You have just identified a point in time at which your
> program could run without problems. We have not even attempted to make a
> fix for this that could be merged.
>
> As you can see in the commit message in the commit that breaks the
> functionality for you, that commit was a partial revert of a previous
> change I did to that code. It thus has an established history of being
> fragile. Let's tread gently.
>
> The short history of this:
>
> cb3e6dfa3511 - I commited a change to Curl_proxyCONNECT to make test 1021
> work with valgrind.
>
> a4cece3d47cf - I modified the previous change since it was pointed out to
> be that the previous fix caused a serious and notable performance
> regression. The fix was confirmed to fix the regression and test 1021 still
> worked.
>
> Today - there's apparently a bug in the code that makes it fail for a
> still not very detailed use case but it involves auth with the proxy,
> CONNECT and then SSL over that. It'd be great if we could get a small
> sample program that can repeat the problem, but I realize it may be tricky.
>
> Reverting a4cece3d47cf will only bring back the performance regression. I
> would instead like us to find what the actual problem is with the current
> code and work on improving it instead.
>
> We're planning the next release to happen November 5. It would be great to
> have a fix done by then.
>
> --
>
> / daniel.haxx.se
> -------------------------------------------------------------------
> List admin: http://cool.haxx.se/list/listinfo/curl-library
> Etiquette: http://curl.haxx.se/mail/etiquette.html

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2014-10-09