The future of portmaster

[Jan Bramkamp]

On 17/02/2017 09:25, Matthieu Volat wrote:
> On Fri, 17 Feb 2017 10:37:16 +0300
> abi <abi at abinet.ru> wrote:
>
>> 17.02.2017 00:22, Chris H пишет:
>>> On Thu, 16 Feb 2017 15:48:57 -0500 Baho Utot <baho-utot at columbus.rr.com> wrote
>>>
>>>> On 02/16/17 15:40, George Mitchell wrote:
>>>>> On 02/16/17 15:33, Baho Utot wrote:
>>>>>>
>>>>>> On 02/16/17 14:01, Lowell Gilbert wrote:
>>>>>>> Baho Utot <baho-utot at columbus.rr.com> writes:
>>>>>>>
>>>>>>>> On 02/16/17 06:08, Luca Pizzamiglio wrote:
>>>>>>>>> I'm looking for constructive critics, feedbacks, anything that can
>>>>>>>>> help me to make portmaster an actively maintained and used tool.
>>>>>>>> If you can have it build in a clean chroot or jail then you'll get my
>>>>>>>> attention
>>>>>>> What kind of special support?
>>>>>>>
>>>>>>> I use it with a chroot that mounts /usr/ports (and src) read-only, and
>>>>>>> aside from the initial base system install, it took about fifteen
>>>>>>> minutes to set up.
>>>>>>>
>>>>>> Using chroot or jails to build each individual package
>>>>>> [...]
>>>>> While I understand the interest in chroot/jails as an optional
>>>>> feature, I hope it doesn't become required.  The current non-use
>>>>> of chroot/jails is, for me, a feature -- not a bug.    -- George
>>>>>
>>>>>
>>>> Having built and packaged linux from scratch using the rpm package
>>>> manager, I came to find that if one is building packages to be used on
>>>> multiple machines, one needs to build each package in a chroot
>>>> environment or the package could inherit things from the parent not
>>>> found in the target machine.  Here by making the package unusable.
>>> Hello. You shouldn't have any difficulty accomplishing your goal
>>> by simply setting up a jail, and using portmaster within that jail(8).
>>> portmaster really doesn't care where it's run. So long as it has
>>> everything it needs to accomplish it's job(s). :-)
>>>
>>  From my point of view, jails are overkill. Chroot should be enough and
>> it would be nice if portmaster starts building in clean environment.
>
> Just dropping privileges to a dedicated user for building would be a big step, but that's more a port feature (openbsd's ports do that, if I'm not wrong).

Yes dropping privileges would be a good *additional* step. The purpose 
of the jail/chroot isn't just for security. The real goal is to provide 
a reproducible, clean build environment. Lots of broken configure 
scripts out there include a lot of autodetection magic. And suddenly 
your binaries are link against additional libraries which are unknown to 
pkg. This becomes even funnier if your application uses fork+exec per 
connection. Suddenly you're left with a bound socket but each connection 
dies because the worker fails to link at runtime. This was the straw 
that broke the camels back for me. An other problem with portmaster is 
that it creates inconsistent during the upgrade by design. Of course pkg 
upgrade isn't atomic either but the time window is on the order of a few 
seconds instead of minutes to hours and is far less likely fail halfway 
through.