cURL / Mailing Lists / curl-library / Single Mail


RE: connect.c patch for progress callback

From: Daniel Stenberg <>
Date: Fri, 8 Jan 2010 11:10:08 +0100 (CET)

On Thu, 7 Jan 2010, Adrian Michel wrote:

> I applied the new patch and here are the results:

As I said, the new patch has no different functionality than the previous one.
The fact that you get different results with this and the previous one implies
that your testing method is unreliable.

> 1. intially the host name was in the hosts file and I was able to connect
> with both versions of the dll (w/ and w/o patch) - expected

Since the patch only modifies the function that waits for the *connect* to
complete it was totally expected, yes. The connect happens _after_ the name
has been resolved properly.

> 2. I removed the host name from hosts and I wasn't able to connect with
> either version (w/ and w/o patch) - expected. BTW, the version with the
> patch returned the error result much faster (almost immediately), while the
> version w/o the patch took a quite long time (around 20 seconds).

So it then resolved the name with DNS and failed to connect to that address
faster with the new approach. Can you find anything that would explain this
behavior? It seems very odd to me.

> 3. I added the host name to hosts, and now the version w/o patch connects
> fine, but the version with the new patch doesn't connect any more.

So you're saying that it worked first when you had it in hosts, and then
stopped working when you had removed the hosts entry and put it back again?
How on earth would libcurl know or care about if you've touched your hosts
file? This doesn't make sense at all.

> Now, no matter whether the hosts contains the host name pointing to the
> local machine or not, the version w/ the patch fails consistently, while the
> version w/o the patch works as expected (other than not being able to get
> the call to the progress callback).

Based on these shaky results I no longer have faith in your tests so I'm not
sure how to proceeed. I think you should stop just explaining what happens and
instead enable CURLOPT_VERBOSE and show us proper libcurl ouputs to describe
what happens or similar. I'm suspecting you suffer from problems elsewhere and
those problems fool you to draw wrong conclusions on these libcurl changes
we're trying out.

With this patch applied, do you see any problems when using (lib)curl against
other sites/URLs ?

List admin:
Received on 2010-01-08