[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: automatic package statistics
Simon Burge <simonb%NetBSD.ORG@localhost> writes:
> Julian Assange wrote:
> > I've never had any feedback on whether people use these or not. I am
> > wondering whether they are worth maintaining (they were certainly a
> > gargantuan effort to create). I feel certain that other pkg authors
> > are also in the same boat. I propose therefor, that bsd.pkg.mk and
> > pkg_add be modified to send pkg.netbsd.org a single udp packet (with
> > uname -a, and package info) under the following circumstances:
> > a) on a successful / unsuccessful distfile/pkg fetch
> > b) on a sucessful / unsuccessful build
> > c) on install / pkg_add
> A couple of comments:
> a) I think that _if_ something like this went ahead, the default
> should be off. Users can be told that if they want to help
> with statistics gathering, they can set a "PKG1984=yes"
> variable in /etc/mk.conf. Otherwise I suspect the backlash
> ("Look what NetBSD is doing behind my back!") might be too
opt-in would make it useless. And the sentiment is irrational. The
way to deal with such irriational sentiment is (a) allow people
to turn it off. (b) explain why it's irrational.
> b) How many people are behind a firewall of some sort? If you
> want any meaningful statistics (or even just lies and damn
> lies), you'd need a more reliable transport than UDP. Email
> comes to mind.
I considering email. The problem is that
(a) email is easy to misconfigure
(b) uses far more disk space, bandwidth and cpu
(c) leaves more logs than it needs to
(d) may pose a significant burden on the receiving machine
(e) is ugly if it ever bounces
We don't need the statistics to be completely accurate over n to
be useful. Just so long as any bias is reasonably uniform.
We can send packets out to port 53 to bypass many firewalls.
Main Index |
Thread Index |