[petsc-dev] I hate nagupgrade

[Sean Farley]

Matthew Knepley writes:

> On Tue, Nov 11, 2014 at 3:55 AM, Jed Brown <jed at jedbrown.org> wrote:
>
>> KAUST (or KSA?) internet can be flaky at times and my "make" was
>> (silently) hanging indefinitely while trying to connect to mcs.anl.gov.
>> Manually touching .nagged allows my build to proceed.  The hang could be
>> fixed by adding a reasonable timeout, but I can't find a timeout in
>> urllib.  Aron suggests that I try curl because all built-in Python url
>> libraries are terrible, but I don't think we can depend on curl being
>> installed, so we'd have to fall back to something.  We could implement a
>> timeout using threads, if threads weren't broken on some architectures.
>>
>> Meanwhile, the professor next to me runs Little Snitch on his Mac and
>> wants to know why PETSc's build is trying to connect to Argonne's
>> servers.  His first thought was that it was a there for surveillance.
>>
>
> I find it hard to believe he is a science professor with that kind of
> inference
> from the data (you can see it retrieves a webpage). Maybe climate ;)
>
>
>> PETSc has a significant number of users that work behind firewalls or
>> are otherwise sensitive to outgoing connections.  Although nagupgrade
>> helps people stay updated and reduces some support email, I think it is
>> unprofessional and a failure mode that I'd rather avoid.
>>
>
> urllib2 has the timeout argument, so we should switch. I am not sure I see
> retrieving a webpage as unprofessional. Is there a better way to update
> information, or do we want a model that is completely dead once downloaded?
> I think people now assume that this is not true.

Speaking from my experience as being a package maintainer, all of 0
packages do this 'phone home' in MacPorts. That's a sample size of
almost 20,000 packages.

One of the first things we package managers would do in this case would
be to remove this "feature," since, as Jed mentions, there are many
valid reasons to not have a connection to mcs.anl.gov (privacy
concerns?). It is most definitely unrealistic to have this kind of
behavior in production.

In general, I would love PETSc to "fall in line" more with a package
manager layout but that's a different thread :-)

Just my two cents, of course.