On Friday 09 December 2005 02:07, Andrew Vaughan wrote:
> On Fri, 9 Dec 2005 05:02, Malcolm Rowe wrote:
> > I've now managed - I think - to reproduce the problem we're seeing using
> > my own 'client' (which is a client of libsvn_fs, but not a 'client' in
> > the normal sense). All it takes is for us to get a transient disk-write
> > error when writing the delta rep, and the caller to _not_ immediately
> > abort the transaction (which according to the documentation for managing
> > transactions in svn_fs.h, is fine).
> >
> > (I can't prove that any of our clients do that, but I can't prove they
> > don't, either). Anyway, at this point, we're probably okay until the
> > pool cleanup runs.
>
> There are multiple other subversion clients. Some of them probably use
> subversion apis in a different manner. I wonder if anything tries to retry
> a failed call without cleaning up properly... maybe re-using the same data
> structs from the failed the original attempt? Or retries when they really
> should abort?
Most clients drive everything through the client APIs, so underneath the hood,
they should all be doing the same thing.
> John, do you know which client was used for the problem commits?
IIRC, all but one used the command line client. The one was used TortoiseSVN.
-John
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Dec 9 09:35:40 2005