| 1 |
On Fri, Aug 1, 2014 at 10:35 AM, hasufell <hasufell@g.o> wrote: |
| 2 |
> Rich Freeman: |
| 3 |
> We only have the current implementation. So what you are saying is again |
| 4 |
> just an idea without code. Voting on it now without a reference |
| 5 |
> implementation will not result in anything useful. |
| 6 |
|
| 7 |
Oh, I agree completely. I was trying to get at that in the last line |
| 8 |
of my email. |
| 9 |
|
| 10 |
>>> * They don't work with overlays. |
| 11 |
>>> |
| 12 |
>>> * They don't work with "resurrecting" packages in overlays. |
| 13 |
>> |
| 14 |
>> A lot of things don't work with overlays when it comes to changing |
| 15 |
>> dependencies. The only reason we can do a lot of things in portage is |
| 16 |
>> that we modify reverse deps along with the deps themselves. |
| 17 |
>> |
| 18 |
>> I haven't thought overlays through, and I'm willing to believe that |
| 19 |
>> things break in odd ways with overlays. We may never be able to |
| 20 |
>> handle dynamic deps properly with them. |
| 21 |
>> |
| 22 |
> |
| 23 |
> Gentoo is becoming anti-modular. That is contradictory to our |
| 24 |
> meta-distribution model. |
| 25 |
|
| 26 |
Most of the overlay issues are with namespace. Overlays work great if |
| 27 |
they are self-contained. If they depend on packages in the main |
| 28 |
repository and those packages change, they will break. |
| 29 |
|
| 30 |
> |
| 31 |
> Not only, but also because of that we will lose a lot of people to NixOS |
| 32 |
> in the next few years, I am pretty sure of that. |
| 33 |
> |
| 34 |
|
| 35 |
I don't see how they would solve the issues we have with overlays. I |
| 36 |
don't know much about them but their approach to isolated builds |
| 37 |
should at least help prevent dependency errors in the first place. If |
| 38 |
package foo splits into two packages with different names, I don't |
| 39 |
know how NixOS would prevent the need to update reverse dependencies |
| 40 |
(unless it expresses things more granularly). |
| 41 |
|
| 42 |
On the topic of dependency verification: one thing that probably |
| 43 |
wouldn't be terribly hard to do is to configure our build jails based |
| 44 |
on the declared DEPENDs. The jail already has both inclusion and |
| 45 |
exclusion functionality - it is just set to include everything that |
| 46 |
isn't explicitly excluded. If you fed the jail a list of everything |
| 47 |
in DEPEND and @system then you should be able to reliably detect |
| 48 |
missing DEPENDs. |
| 49 |
|
| 50 |
But, that is more non-existent code. |
| 51 |
|
| 52 |
What might be another approach would be to actually containerize or |
| 53 |
use SELinux to enforce RDEPENDs when a package runs. I'm not sure if |
| 54 |
NixOS purports to do that. |
| 55 |
|
| 56 |
Rich |
Archives