Re: [PATCH] "svn commit" performance

Chia-liang Kao <clkao@clkao.org> writes:

> the problem (and the sort) here is that if foo/bar/baz is locked first,
> subsequent tree locking of foo/bar would fail since baz is already locked.

Ah, yes that would be a problem. In fact I mentioned the problem of
already locked but non-recursive baton in my last mail but I didn't
realize how serious it was, it really does need to be fixed.

I feel that having a requirement to sort the paths first means the
that the svn_wc_adm_open_descendant API is broken, the user has to
know the entire history of the access baton, which conflicts with the
"open if not already open" behaviour of this function. It really
would be better to get it to do the right thing, possibly by
refactoring the tree_lock stuff from do_open. I suspect that doing
this would also get rid of the closing and re-opening of the basedir
problem.

-- 
Philip Martin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org