>
> I beleive my change is more correct. Given the sparse use of the
> "record" functions this change will not yield a fix to any existing
> problem. However, it will be more correct for future use.
> I can submit a real patch if you like.
Not! I have a little problem with my logic also.
The checkpoint bothers me. It is possible, "reading bdb docs" that a
checkpoint could fail after a commit succeeded.
This is not goodness. This would/could create an unexpected undo behavior.
Two things in particular:
1) svn_fs__record_completion recorded functions could get called twice.
2) svn_fs__record_undo recorded functions would be called even though the
commit succeeded.
As I understand it. Even if the checkpoint fails the commit information has
not been lost. Oncde the recovery is run it should be there. Right?
So, I think the checkpoint should be run after the commit_trail function
returns.
This is the way I'm coding it in my new version.
It will be called as part of an internal flush_trail function.
cheerio,
gat
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Oct 17 00:49:56 2002