Dean Roehrich <Dean.Roehrich at sun.com> writes: > On Wed, Jul 09, 2008 at 09:58:04AM -0700, Stephen Hahn wrote: >> * James Carlson <james.d.carlson at sun.com> [2008-07-08 13:57]: >> > That means that we really do need to stop usage of /ws/onnv-gate as a >> > gate, and use hg the way it was meant to be used -- with >> > ssh://onnv.sfbay/ instead. >> >> Yes. >> >> > If /ws/onnv-gate exists at all, it needs to be just a copy of the >> > source code itself, rather than an actual Mercurial repository. (I >> > know some people cd over to the gate and run cscope; that usage should >> > probably be preserved.) >> >> It could continue to be shared read-only; I don't believe we have any > > The repo that receives the pushes should not be available to anyone via NFS. > It would have to be a downstream clone that is made available for your > community-cscope purposes. Also, you'd need to add a gate hook that does an > update after each push completes...ug, don't.
That has to be the case anyway, the main onnv-gate is going to be on opensolaris.org. :) (well, except for usr/closed). So I agree, this has to be planned with the assumption that what we make available from onnv.eng is not *the* gate, except for usr/closed, but a (possibly fast moving) clone thereof. > A pull can pick up changesets that have been pushed, but haven't had their > transaction closed. Then the gate hook might decide to abort the transaction. > Now you have someone out there who (inadvertently) snuck in and obtained a > copy of the now-rejected changesets, and this will, eventually, lead to a > point where things get very confusing as those changesets continue to float > around among promiscuous users. > > Rich had been talking about adding a lock to the gate push hook, and a pull > hook that would check that lock...dunno. If pulling via NFS, would it be > possible to prevent that pull hook from firing, or from working properly? I don't know, right now. I didn't get chance to check this out, what with everything else to do. -- Rich
