[Popcon-developers] Bug#496831: Detecting data package usage

[Amos Jeffries]

Bill Allombert wrote:
> On Sat, Jul 25, 2009 at 05:24:31PM +1200, Amos Jeffries wrote:
>> I also am trying to keep track of several similar packages with  
>> data-only contents.
>>
>> Perhapse the fix for this would be a package tag maintainers could add  
>> that specify a file for popcon to test. Sort of like a "use details  
>> available about this file if it exists".
> 
> I do not like the idea of allowing package maintainers to tweak packages to
> improve their ranking.

I've missed something here. If the file is required to be part of the 
installed package how can the maintainer abuse the stats to get more 
votes than installed?

Or are you only referring to the 'recent' stats?  I would think 
maintainers have more reason to keep those accurate and know the speed 
of upgrades than to fake and make recent==installed.

As it stands it's trivial for maintainers to add a cronjob to their 
package and touch the binary every minute or so should they want to ruin 
their stats.

If they use the proposed tag properly they can point it at a data file 
and get realistic stats about how often its used. Right now we have a 
situation where installed == nofiles == N% and recent == vote == 0 for a 
lot of packages which no only contain files but _are_ used and upgraded 
over time.

You could go through yourselves and alter popcon for each and every one 
of those packages, or add a way for maintainers to quickly give you a 
hand with the classification.

> 
>> Or if that is not possile, defaulting to checking the changelog. (which  
>> every package has right?)
> 
> What would be the point ? There is nearly zero correlation between the
> time someone read the changelog and someone use a package, *especially*
> for data-only packages.

I was thinking for installed ctime and updated mtime stats.
I agree atime for use is not available if such a fallback was needed.

AYJ