can you help collect package popularity data?

[Tim Abell]

Alfa21 wrote:
> 2010-04-10 at 23:36 Tim Abell
>
>   
>> The code is at http://github.com/timabell/popularity
>> improvements would be welcome, as would inclusion in the various distros.
>> Perhaps if it works out this could be hosted by the various distro servers.
>>     
>
> nice idea...
> but it works only with opkg based distros and only with _packages_ (no manually installed tgz or compliled sources)
>
>   
i'm only interested in opkg distros (shr) at the moment. apt bases ones 
could presumably use popcon if they want this functionality. tracking 
multiple distros would be a much more ambitious project, needing more 
standardised communications, and a way of matching up varied package 
name conventions to actual software. You are welcome to have a go :-) 
One that tracks software accross *all* platforms (including windows) 
would certainly be awesome for the OSS community.

I don't think I mind missing other methods of installation, as generally 
all programs that see any decent usage *should* be packaged properly 
(it's very easy to provide an ipk, even if you don't put it any feeds). 
Besides, the itch I was scratching was for an app that *is* packaged <grin>.

I think it would be harder to figure out what has been installed from 
source & tgz, they can be too varied to track. (patches welcome though :-)
> also, it seems you do not do any check on duplicated submissions from the same user.
>
>   
indeed, I'd need a unique id for that. the trade off for that is 
privacy. not that i'm particulary against it, I just wanted "some 
numbers" as opposed to "no numbers", so wanted to get something out 
there without trying to make it perfect (and therefore never releasing it)
> so I think a human submitted simple "poll" is more useful.
>   
i think human polls have their own value, but I think you are more 
likely to get complete data from something more automated. for what I 
wanted, the programmatic approach suits best.
> fixed choices with the label of the application (imho version number is not strictly relevant)
version is part of the default output of "opkg list-installed". I 
stripped it out when processing the data to just get a count & package 
name. see http://www.timwise.co.uk/om/pop/summary20100410.txt for an 
example. Version number might be useful to see if people are running 
out-dated / insecure versions, prompting a push to get everyone upgraded.
>  and the possibility for any user to add a new slot for missing ones and when submitted these new slots are converted to fixed choices for the others.
> every user should have own identifier so it's possible to update the own list of installed stuff.
>   
it would be good for a user to have their own unique id that they could 
enter as well as a per machine id, but it would have to be optional as 
it might put people off. better data vs. privacy again I think.
> the date of submission should be collected too.
>   
the filenames contain the php date output, which presumably can be 
turned back into a meaningful value.
> and ask a captcha on submission (bots are everywhere!)
>   
I'll deal with that when I have a problem with it ;-)

Thanks for your email, it's great to bounce these ideas around.

Yours

Tim Abell