[MERGE] Fix bug #225020 by catching CURLE_SEND_ERROR error.

[Robert Collins]

On Thu, 2008-08-28 at 09:12 +0200, Vincent Ladeuil wrote:
> 
> 
> This patch fixes the problem encountered when running the test
> suite with pycurl installed.
> 
> It appears that a recent pycurl/curl modification have introduced
> a bug in pretty rare occurrences.
> 
> I don't have the specific versions but since I never experienced
> it while running gutsy and ran into it almost immediately after
> upgrading to hardy, we could find and fix it in pycurl/curl if
> needed.
> 
> Instead, I chose to catch the error in our code because it was
> simpler and didn't seem to have drawbacks.
> 
> The bug occurs when we probe an HTTP/1.0 server for a smart
> server.
> 
> The server replies with a 403 error and close the connection
> *but* pycurl still try to send the body request, on a closed
> socket, and fails.
> 
> Most of the time though, it appears that pycurl/curl is able to
> send the body *before* the connection is closed avoiding the bug.
> 
> This seems to occur more often in the test suite since a single
> process is used for both the client and the server.
> 
> Trying to reproduce the problem with a dedicated test looks like
> overkill when I had already 35 tests failing.

bb:approve

FWIW I think its a change in socket behaviour in the kernel, that also
made bzr's minor idiosyncracies in  1.5ish make hpss really slow (until
we fixed them). So likely a very old curl bug :). Consider filing a bug
upstream too please.

-Rob
-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20080828/68469693/attachment.pgp