On Sun, 12 Feb 2006, Shmulik Regev wrote:

> My problem was that the request was SUCCESSFUL at least HTTP-wise. libcurl
> had no way of knowing the wrong server was picking up the request (due to
> some DNS tricks) and the connection persisted (such were my settings).
> However, the problem (as I see it) is that Curl_resolv_unlock (called by
> Curl_done) doesn't attempt to prune the cache.

I'm not so sure pruning should be done in unlock, since pruning can be quite
costly and there can be several unlocks done during a single request. I'd
rather try to limit it to just one pruning when the request is complete. Like
adding it in Curl_done() after the possible unlock has been done.

That should work equally fine for your case, shouldn't it?

> Thus adding a final statement to Curl_hostcache_prune should fix this issue.

I take it you meant Curl_resolv_unlock.

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html