Aaron,
I think this is either a bug in the tig_r poll call, or in the apr_poll
implementation. Near as I can tell, this occurs when poll falsely
returns a result indicating that input is available. But I hadn't seen
this issue on Panther, so I'm a bit surprised by that. I filed a bug
with Apple this afternoon (in which I claimed this was probably a tig_r
bug).
I don't think the server side has anything to do with the issue: the
server in my case is a system running panther.
At this point I think there's either a bug in apr, or in the underlying
poll or select code. Can you confirm for sure that you've seen this
same behavior on Panther?
(By the way, I've been able to eliminate the problem [and get back to
work] by simply returning 0 out of the routine
svn_ra_svn__input_waiting in libsvn_ra_svn/marshal.c). Near as I can
tell, that routine is only called defensively to verify that the
protocol is working properly. It's unnecessary for proper operation of
the protocol. This whole problem is a result of debug code that's
running afoul of a bug.)
Which isn't to say, of course, that we shouldn't help somebody track
down the underlying bug. As I said above, this is either a bug in the
system poll or select code, or a bug in the apr_poll routines. Note
that as-of apr 1.0 or 1.1, there's an apr_poll implementation on
Panther+ based on kqueue, which may not exhibit this problem.
If you have any further data or insights please let me know and I'll
add them to my bug report. If you can confirm for sure that this
manifests on Panther, that would also be an important data point.
-jdb
On Feb 14, 2005, at 4:05 PM, Aaron Burghardt wrote:
> On Feb 13, 2005, at 4:25 PM, James Berry wrote:
>> On Feb 13, 2005, at 8:49 AM, Ben Collins-Sussman wrote:
>>> On Feb 13, 2005, at 10:35 AM, James Berry wrote:
>>>> Can somebody give me a clue to what I can do to stop this problem,
>>>> or where I might start looking to determine the source of this bug?
>>>> Would a dump of the tcp stream help?
>> TCP flow traces and tcpdump output are here:
>> http://www.jberry.us/~jberry/svnbug/
>> Let me know if you need some other format.
>> The only other things I can think of that are a bit weird about my
>> installation is that I'm running Mac OS X Tig_r on the client side,
>> and the client sw is built in that environment with gcc 4. Might this
>> be a bug in apr and/or with Tig_r?
>> -jdb
> I am ready to report exactly the same problem. I have produced the
> error on OS X Panther and Tig_r with both 1.0.6 and 1.1.2/1.1.3 when
> connecting via svn:// to a Windows 2000 server. I initially had a BDB
> repository running Subversion 1.0.6 on the client and the server. I
> tried upgrading to 1.1.3 on the server and 1.1.2 on the client (using
> Fred Sanchez's packages for both the 1.0.6 and 1.1.2 clients). I also
> tried changing the repository from BDB to FSFS with a dump/restore.
> I can get really small commits, 2 to 4 files, to complete, but more
> files than that tend to fail.
> What other detail would be useful?
> Thanks,
> Aaron
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: users-help@subversion.tigris.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Tue Feb 15 02:51:49 2005