On Fri, 16 Nov 2007, Bill Nottingham wrote:
Nicolas Mailhot (nicolas.mailhot(a)laposte.net) said:
>> scriptlets should be allowed to fail when the failure is catastrophic
>> enough. What that is, I'm not sure.
>
> scriptlets should be allowed to fail when the benefits of failing
> (fixing packages) outweigh the cost of failure (killing transactions for
> lots of users). Since so far the only documented failure was in rawhide
> (at a time I doubt rawhide was perfectly installable anyway) I question
> the need to hide this particular failure.
Not all scriptlets abort the transaction anyways, as I recall. %post
certainly doesn't. So, maybe it should just be %pre shouldn't fail.
No scriptlet will abort the transaction, just the package with failing
scriptlet will have its install/upgrade/remove will be aborted.
"We brake for nobody" ;)
- Panu -