On 4/22/2016 4:16 PM, BoBo via curl-library wrote:
Hi.
Using the same handle I want to download the same URL (e.g.
microsoft.com <http://www.microsoft.com/>) first time without proxy
and second time with wrong proxy.
In order to achieve that I am setting proxy to empty string before the
first download:
curl_easy_setopt(handle, CURLOPT_PROXY, "");
curl_easy_perform(handle);
And setting it to something wrong before the second download:
curl_easy_setopt(handle, CURLOPT_PROXY, "wrong:80");
curl_easy_perform(handle);
The expected correct result would be to succeed for the first time,
and to fail for the second time, because the second download is using
wrong proxy.
But the actual result I am seeing is successful downloads for both
cases. So I have the impression that although I am changing proxy
option before triggering second transfer, the second transfer reuses
the successfully established connection (one without proxy) instead of
trying to create a new connection with new proxy information.
I have also tried with CURLOPT_FRESH_CONNECT and CURLOPT_FORBID_REUSE
options but didn't observe any change in
behavior.<http://stackoverflow.com/questions/36730394/libcurl-disabling-proxy-when-using-same-handle#comment61044727_36730394>
Two observations that I have made are:
1. If instead of immediately triggering the second download I wait for
2 minutes and only then trigger the second download with wrong proxy,
the behavior is as expected (the second download fails).
2. This issue occurs only when my wrong proxy address has port 80
(somewrongproxy:80). If I specify wrong proxy with port other than 80,
everything works as expected.
So can you please help me figure out how is port 80 special in this
case and how I can reuse same libcurl handle so that proxy changes
between transfers take effect.
curl -v http://test.com --next -x wrong:80 http://test.com
curl -v http://test.com --next -x wrong:81 http://test.com
I think the reason it fails when 81 is because it's looking in the DNS
cache for test.com port 81 and sees it doesn't exist whereas test.com
port 80 exists.
CURLM_STATE_WAITRESOLVE in the state machine has [1]:
dns = Curl_fetch_addr(conn, conn->host.name, (int)conn->port);
But shouldn't that be conn->remote_port ?
And none of this explains why for wrong:80 it connects to test.com:80 as
if it's the proxy.
[1]: https://github.com/curl/curl/blob/curl-7_48_0/lib/multi.c#L1168-L1175
-------------------------------------------------------------------
List admin: https://cool.haxx.se/list/listinfo/curl-library
Etiquette: https://curl.haxx.se/mail/etiquette.html
