* Eric Schrock <eric.schrock at sun.com> [2008-07-16 23:18]: > On Wed, Jul 16, 2008 at 03:56:06PM -0700, Stephen Hahn wrote: > > > > I believe we're in a better understood situation than these paragraphs > > suggest. > > > > For SSH, you can either use a captive shell environment, as Bill > > writes and is already implemented for hosting on opensolaris.org, or > > you can use a shared account, by populating it with the keys of your > > approved committers. (NFS commits are already not permitted in the > > current TeamWare configuration.) Which we choose for the temporary > > period where the ON gate is still on SWAN is the decision. > > > > The fact that this is a temporary situation doesn't seem to avoid the > fact that such a decision needs to be made before snv_97 opens, unless > we accept that executing any hooks will be optional during this period. > And presumably the situation is not temporary for the closed gate, which > will continue to exist only on SWAN? Yes. My recommendation would be a shared account, having constructed and operated a captive shell environment for opensolaris.org. The drawbacks of that particular captive shell environment are complexity and the loss of interactive login to a system. The latter seems more jarring than having to manage a shared file of public keys.
> To me, the current situation doesn't seem tenable, and no one has > described any concrete plan to rectify the situation before the > transition on August 5th. That plan could "ignore the error message and > please don't do anything naughty", but it would be nice to have that as > a written policy somewhere. That's a reasonable fallback, but I think we can settle on a specific deployment choice. > P.S. This is also not just about commits. This error message is output > on read-only operations like 'hg status'. While we may not have any > hooks that run during such operations, the error message is identical > and its ramifications (did I just miss some important hook?) are not > clear. One option is to have a child (like the current clone) updated as part of the gate hooks; that child could explicitly not an hgrc of its own. - Stephen -- sch at sun.com http://blogs.sun.com/sch/
