Re: [gentoo-dev] Tinderboxing efforts in Gentoo

[Jorge Manuel B. S. Vicetto]

On Sat, 3 Dec 2016, Rich Freeman wrote:

> On Sat, Dec 3, 2016 at 9:47 AM, William L. Thomson Jr.
> <wlt-ml@××××××.com> wrote:
>> On Saturday, December 3, 2016 9:33:00 AM EST Michael Orlitzky wrote:
>>> On 12/03/2016 09:25 AM, William L. Thomson Jr. wrote:
>>>>> This is generally considered infeasible:
>>>> I would not think such, just need a wrapper to run around each package
>>>> that
>>>> would get its USE flags and re-emerge it a few times.
>>>
>>> If a package has 10 USE flags, and if each can be set on/off with no
>>> constraints, then that gives you 2^10 different ways to emerge it.
>>
>> May make the requirements of the host system larger or take more time.  I am
>> sure processing power could handle that load.
>>
>> Would be nice if someone like Google would sponsor such efforts. They have
>> enough hardware and cloud services to make such feasible.
>>
>
> Have you given thought to how long it would take the largest
> supercomputer in the world to rebuild libreoffice once for each of the
> 2^28 USE flag combinations it has (not including USE_EXPAND)?
>
> It is certainly possible, but I doubt that you're going to get it
> dedicated to Gentoo for a few weeks whenever one of its deps changes.
>
> This is not a reason to give up on tinderboxing.  This is just a
> reason to be realistic about just what it will do.

I recall some discussion about tinderboxing suggesting that more important 
than testing all possible use flag combinations, would be to test:

  * all use flags disabled (some packages fail miserabily if you don't 
enable at least one use flag)
  * all use flags enabled (good to detect conflicts)
  * default use flags (rely on IUSE defaults - even better if you test it 
on different profiles to see the impact)
  * a set of use flags defined by maintainer (mysql ebuilds have a set of 
USE flags for maintainers to test them)

Regards,
Jorge Manuel B. S. Vicetto
Gentoo Developer