Review queue/FESCo after the merge

[Panu Matilainen]

On Fri, 16 Nov 2007, Bill Nottingham wrote:

> Matthias Clasen (mclasen at redhat.com) said:
>>> Yes, but fixing every possible application which could ever be put in a
>>> scriplet to never, ever, ever fail, even in the cases when failure is
>>> acceptable (and shouldn't kill the transaction) is a little beyond the
>>> mandate of the FPC. ;)
>>
>> If scriptlet are not allowed to ever, ever, fail, then just make rpm
>> ignore the exit code of scriptlets.
>
> scriptlets should be allowed to fail when the failure is catastrophic
> enough. What that is, I'm not sure.

Agreed on principle but... The rpm transaction is not unlike a derailed 
train - a few red lights from failing scriptlets ain't going to stop it 
from wrecking everything that happens to be on its way until it simply 
runs out of speed. So the problem is: what exactly is a scriptlet 
intentionally failing going to do? It wont stop the transaction anyway...

For the vast majority of cases it'd be far far more useful just to 
ignore the status but log the failures to permanent storage (and notify 
user at end of transaction). Leaving duplicates behind on upgrades like it 
now does is hardly useful behavior to anybody.

Then of course there are packages (typically commercial) that attempt to 
prevent installation by non-zero exit from %pre if some conditions aren't 
met, but that's totally broken to begin with because it wont prevent the 
rest of the transaction from running...

 	- Panu -