[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: Release build serf test failures

From: Paul Burba <ptburba_at_gmail.com>
Date: Tue, 3 Nov 2009 16:29:43 -0500

On Tue, Nov 3, 2009 at 2:12 PM, Lieven Govaerts <svnlgo_at_mobsol.be> wrote:
> On Tue, Nov 3, 2009 at 7:00 PM, Paul Burba <ptburba_at_gmail.com> wrote:
> [..]
>> I cleared the access and error logs, restarted the server and ran just
>> merge_tests.py 78.  The test run output and the two logs are attached.
>
> It looks like the client is doing a get_location_segments action and
> the resulting REPORT request returned successfully from the server. At
> least, I assume this is the last REPORT in the access.log file
> (revision 7 matches the start_red and end_rev vars in the stacktrace).
>
> I can't see what's going wrong.
>
> Maybe you can abort the test right before the merge and run it
> manually, with some additional printf's? My first guess would be to
> look in util.c:handle_response right before and after the call to
> ctx->response_handler.

I've done a lot of printf debugging already, the problem occurs before

1346: err = ctx->response_handler(request,response,
ctx->response_baton, pool);

Rather it is back on

1228: status = serf_bucket_response_wait_for_headers(response);

Where the problem manifests itself. The status comes back as 70014
(APR_EOF) which causes us to hit

1244: if (strcmp(ctx->method, "HEAD") != 0 && sl.code != 204 &&
sl.code != 304)

Which segfaults because ctx isn't valid. The debug build *never*
returns APR_EOF on line 1228 so that avoids the segfault.

All the other failing tests besides #78 have a valid ctx so we enter
the block after line 1244 and see the 'svn: Premature EOF seen from
server (http status=404)' error.

------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2414221
Received on 2009-11-03 22:30:00 CET

This is an archived mail posted to the Subversion Dev mailing list.