On Fri, 2006-03-03 at 13:06 -0800, David Lutterkort wrote: > On Fri, 2006-03-03 at 14:37 -0500, seth vidal wrote: > > define a failure for me: > > - if you ask to install 5 packages and 2 of them are not available in a > > repo and 1 of them is already installed but 2 of them can be installed - > > is that a failure or a success? > > > > - if you ask to remove 2 packages and 1 is not installed and 1 is > > installed. Is that a failure or a success. > > Good questions. I would base the definition of failure on a reasonable > definition of what the state of a system should be after running the > corresponding yum command (and I am sure others will disagree ;) If the > system is in any other state after the yum run, be that because a > dependency doesn't exist in a repo, the network is down, ... that should > be flagged as a failure witha non-zero exit code. > > If I run 'yum install p1 p2 ...' I would define success as all the > packages being installed on the system after yum finishes. > > If I run 'yum remove p1 p2 ...' success should be that none of p1, > p2, ... are installed on the system after yum finishes. but it's not just discrete package names, either: yum install p1* p2* for a long while we questioned whether or not: yum install p1 p2 when p2 is already installed shouldn't be an error, if only b/c it is useful to know that for dealing with that from a shell script. and trivial operations like: yum list foo if foo didn't exist some folks wanted it to error out (I'm not joking) Finally we are already using non bitmask result codes with yum check-update and a locking failure. We can definitely implement some sort of error code but we need to be fairly strict about it lest we waltz into something horrible to figure out. -sv
