autoinstallers should not assume locker 'foo' mounts at '/mit/foo'

[mitchb]

The autoinstallers make an invalid assumption that if you 'add foo', it lands at '/mit/foo'. This is not necessarily true if you obey the hesiod filsys pointer's mountpoint field, as Athena 9 does. Athena supports alias-type lockers (see, for example, 'isng' which is an alias for 'sppin', and RT ticket 878251).

If you have an automounter, you can cd into the /mit/alias and have it work, but on both Athena 9 and Debathena, 'add alias' will actually only create '/mit/realtarget'. So the autoinstallers should look up and use the hesiod-specified mountpoint.

[ezyang]

A possible solution, suggested by geofft, for this is to switch /mit to use pyhesiodfs, or use autofs. I, for one, like fixing our autoinstalls, but we should libify whatever code we use to get the full path (if it's not trivial). I don't know what the hesiod command is off hand.

[mitchb]

No, the idea you mention is invalid and misunderstands the problem. You get no pyhesiodfs or autofs on Athena 9. We will indeed need to fix the autoinstallers to use hesiod, as originally stated. This is trivial, and just requires a few cycles to find all the places we need to do it, and test that it still works correctly everywhere.