FSFS corrupted revfiles && APR minimal version?

I just checked the APR site and the latest versions are
1.2.11 and
0.9.16. While checking their CHANGES files (respectively at
http:/
/www.apache.org/dist/apr/CHANGES-APR-1.2 and
http:/
/www.apache.org/dist/apr/CHANGES-APR-0.9), I noticed
that there
are some fixes which are at least interesting for us. My
question is:
Could these be related to the FSFS revfile corruptions we
have seen
and if so, should we up our requirements?

1) "Fix file pointer position calculation in
apr_file_writev()." (apr
0.9.14 and 1.2.9)
2) "Fix thread safety of assorted file_io
functions" (apr 0.9.14 and 1.2.9)


Does anybody have an answer to these questions?

bye,


Erik.

PS:


Next to the 2 above, there's one for OS/X which we might
want to
require to prevent BFI (Bad First Impression):

- "Define OS/X filesystem encoding as UTF-8" (apr
0.9.15 and 1.2.10)

------------------------------------------------------------
---------
To unsubscribe, e-mail: dev-unsubscribesubversion.tigris.org
For additional commands, e-mail: dev-helpsubversion.tigris.org

Erik Huelsmann wrote:
> I just checked the APR site and the latest versions are
1.2.11 and
> 0.9.16. While checking their CHANGES files
(respectively at
> http:/
/www.apache.org/dist/apr/CHANGES-APR-1.2 and
> http:/
/www.apache.org/dist/apr/CHANGES-APR-0.9), I noticed
that there
> are some fixes which are at least interesting for us.
My question is:
> Could these be related to the FSFS revfile corruptions
we have seen
> and if so, should we up our requirements?
> 
> 1) "Fix file pointer position calculation in
apr_file_writev()." (apr
> 0.9.14 and 1.2.9)
> 2) "Fix thread safety of assorted file_io
functions" (apr 0.9.14 and 1.2.9)
> 
> 
> Does anybody have an answer to these questions?

I'm not sure I can really answer these questions, but I'll
try.

1) apr_file_writev doesn't appear to be used by SVN (a least
that's what 
a quick grep has revealed).

2) I'm not sure these matter either.

The root of the problem seems to stem around the fact that a
network 
session can end (from some sort of error), but the
associated pools 
aren't cleared before the client retries.  The end result is
that the 
client is able to start writing new data to the file (with a
new file 
object), and when the pool is finally cleaned up from the
old session we 
end up with a corrupted rev (because data is flushed into
the file).

So the problem seems to be ours and not APR's.

-John

------------------------------------------------------------
---------
To unsubscribe, e-mail: dev-unsubscribesubversion.tigris.org
For additional commands, e-mail: dev-helpsubversion.tigris.org