1 |
On Sun, 2020-01-05 at 07:30 +0000, Robin H. Johnson wrote: |
2 |
> On Sun, Jan 05, 2020 at 07:27:38AM +0100, Michał Górny wrote: |
3 |
> > Introduce a new eclass that contains common logic needed to test |
4 |
> > and install distribution kernels. This is the eclass common both |
5 |
> > to kernels built from source and installed from binary packages. |
6 |
> I'm 100% in favour of the concept here, but I have some implementation |
7 |
> concerns. |
8 |
> |
9 |
> For the (public) record, Infra has been wrapping genkernel to build |
10 |
> binary kernel + initramfs for internal usage for 5 years now. |
11 |
> |
12 |
> Conceptually, can you clarify why you are depending on dracut here, |
13 |
> rather than using genkernel's initramfs mode? |
14 |
|
15 |
For a start, because dracut is not bundling all its dependencies. It's |
16 |
also tested outside Gentoo, so it has much wider testing than genkernel, |
17 |
and doesn't suffer from intermediate deaths, -next forks and total |
18 |
confusion over which version should be used. |
19 |
|
20 |
> I'm not sure you're going to get correct build output in cases like |
21 |
> x86_64 -> x86 cross-compiles. |
22 |
|
23 |
Cross-compile support is untested, and provided merely where it was |
24 |
clear how to do it. I'm happy to fix it later, when someone figures it |
25 |
out. |
26 |
|
27 |
> Why not call genkernel from the ebuild instead of re-inventing the |
28 |
> wheel to compile the kernel yourself. |
29 |
|
30 |
Have you read the logic and found a way to achieve all that from |
31 |
genkernel, without 1) losing advantages of my approach, 2) adding a huge |
32 |
dependency to replace 5 lines of code? |
33 |
|
34 |
-- |
35 |
Best regards, |
36 |
Michał Górny |