To Require yelp or not to require yelp

[Matej Cepl]

On 2007-06-10, 12:19 GMT, Jesse Keating wrote:
> How does yum/rpm know what functionality yelp provides to these 
> packages?  Do we then have to list in the Requires (and auto 
> requires?!) what each required package does for a given 
> package?  Even if the user saw that this would remove yelp from 
> all the packages, and he said yes anyway, how does that get 
> yelp back for them in your firefox-32 scenario?

Well, the difference would be that there would be a big noise for 
the system administrator, so he would know that something 
significant is going on (major functionality of 96 packages 
affected; just length of the list would make me think again).

And no, in my scenario they wouldn't know WHAT EXACTLY will 
happen to them, just that some significant functionality of the 
packages will be affected (because broken soft-dependency 
Recommends:).

Actually, even what you seem to be suggesting was proposed in 
Debian <http://tinyurl.com/2vmkrc> so than you would have in spec 
file something like (quoting the Debian list post):

    Package: mutt
    Suggests: ispell [adds spell cheking while composing emails]
    Suggests: urlview [extracts urls from email and can lanuch 
        a web browser]
    Suggests: mixmaster [allows you to compose anonymized email] 

Why not?

But note that this IS NOT what I suggested. My example would be:

    Package: epiphany
    Recommends: yelp

Then yum would know that when it removes yelp, some (unknown to 
yum) functionality of epiphany package will be affected, and it 
may ask system administrator something in the line of ``Are you 
sure?'' question. And if confirmed (or if these Recommends: 
confirmations would be switched off in /etc/yum.conf) then it 
would just go ahead and remove yelp.

We cannot avoid stupid decisions (and we probably even shouldn't 
try), but IMHO we could (and we should) try to avoid uninformed 
decisions and we should follow the path of the least surprise -- 
when something significant is going to happen in the system, then 
there should be a bang big enough for sysadmin to notice.

Does it make more sense?

Matej