On Sun, 16 Nov 2014 19:26:36 +0100 (CET)
Daniel Stenberg <daniel_at_haxx.se> wrote:

> This patch is still for discussion purposes, it is not complete and
> it causes two test case failures (1208 and 1513) for me.
>
> Thoughts?

+ data->result = CURLE_OK;
+

I think that it will just reset errors that need to be processed.

For example... at the end we have:
         data->result = CURLE_ABORTED_BY_CALLBACK;
 [...]
         result = CURLM_CALL_MULTI_PERFORM;

which reenters this function in order to go to
CURLM_STATE_DONE (or COMPLETED). If that then erases
data->result then you have a serious failure.

+ if(!data->result)
+

without the first line (of which I think is wrong) this would be even
worse of course. But I understand the idea ;).

What about the attached patch? It has no testsuite failures.

-- 
Carlo Wood <carlo_at_alinoe.com>