Re: [DPorts] The only packages available are for DragonFly 3.4

[John Marino <dragonflybsd@xxxxxxxxx>]

On 7/25/2013 15:42, Justin Sherrill wrote:
> On Thu, Jul 25, 2013 at 9:24 AM, John Marino <dragonflybsd@marino.st> wrote:
> 
>> Flip the question.
>> What is so important about this update that it has to published
>> immediately?  Especially when we were publishing only twice a year
>> previously?
> 
> Twice a year?  Do you mean DragonFly releases?  I updated pkgsrc
> whenever I could - doing it on a timed basis was difficult because of
> the packages that would hang and never complete the process for
> upload, but it sounds like poudriere can handle that much better than
> pbulk.

Yes, I mean releases.
Also recall this is exactly what pkgsrc "quarterlies" are: a set of
packages known to build together.  We just do it every 2-4 weeks instead
of every 3 months.  Even if you rebuild quarterlies, you generally don't
get new versions, only patched same versions.  So twice a year is not an
unfair categorization (and the pbulk difficulties help explain that
frequency)

> 
>> Those 50 packages are reverse dependencies, which have exponential
>> ramifications  One package can easily prevent 10,000 or more packages
>> from building.  It's just the way it is.
> 
> I think part of the reason I'm asking this is because I don't know
> where the threshold is for build/don't build.  We know the
> dependencies and what they need to build; is there an acceptable level
> of not-building packages where we can still build newer binaries
> automatically for the packages that aren't affected?
> 
> If we can update 10,000 packages from a build kicked off through cron
> while we figure out while the other 10,000 didn't build, that's
> worthwhile.

I don't think it is.
Right now we're rebuilding 75% of the packages in a 3-4 days.  Once
dillon gets the blade servers running, we'll be looking at building that
in 24-36 hours.  So after a day, you get the report, you fix the
packages and do a delta run or two.  You can probably merge the whole
thing before 4 days passes.

So skipping the whole "how do we mathematically determine which packages
are safe and ONLY transfer those" issue, doing all of this to save 2
days is not worthwhile in my opinion.

John