1 |
On Thu, 18 Mar 2010 23:00:56 +0200 |
2 |
Petteri Räty <betelgeuse@g.o> wrote: |
3 |
> >> Here's how we could change Portage behavior for pulling new slots |
4 |
> >> that are not strictly required: |
5 |
> >> |
6 |
> >> - for packages in the world file install as soon as available |
7 |
> >> |
8 |
> >> - for dependencies install the new slot if everything works with |
9 |
> >> the new slot |
10 |
> >> |
11 |
> >> This would mean that Portage would stay with 2.6 as long as you |
12 |
> >> have something that doesn't work with 3.x installed. |
13 |
> > |
14 |
> > How do you detect this? |
15 |
> |
16 |
> By looking at the dependency graph? |
17 |
|
18 |
But you can't tell whether everything will work with the new slot until |
19 |
you've generated a full set of decisions, and you can't generate a full |
20 |
set of decisions until you decide whether you want to install the newer |
21 |
slot. |
22 |
|
23 |
The problem with expecting the resolver to be clever is that the same |
24 |
kind of clever in different places leads to horrible screwups... Every |
25 |
time the resolver has to make some kind of decision that isn't utterly |
26 |
explicit it's going to do the wrong thing in an annoying minority of |
27 |
cases. Much better to just have ebuilds say exactly what they mean. |
28 |
|
29 |
-- |
30 |
Ciaran McCreesh |