Philip Martin <philip@codematters.co.uk> writes:
> sussman@tigris.org writes:
>
> > This fixes the a real bug in the commit system, as reported by Peter
> > Schuller ("got copy_path, but no copy_rev"). But it also reveals a
> > philosophical bug in our new access-baton stuff.
> > svn_wc_process_committed is unable to recurse on a copied subtree,
> > because children have already been locked earlier by the commit
> > system.
>
> Or it reveals a philosophical bug in the old commit system, it depends
> on your point of view :)
>
> Post-commit processing requires more locks than pre-commit processing.
> This is not a good idea, as the commit might suceed on the server and
> fail in svn_wc_process_committed due to a lock left over from a
> previous command that failed. I think that *all* the required locks
> should be acquired before starting a commit, and then
> svn_wc_process_committed should simply use its given locks, and not
> take new ones.
Totally agree. This new system is much safer! Sorry for
misunderstanding how the old system worked.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Jul 15 20:47:50 2002