1 |
On Thu, 3 Sep 2020 14:49:25 +0100 |
2 |
Michael 'veremitz' Everitt <gentoo@×××××××.xyz> wrote: |
3 |
|
4 |
> I know this is a much more onerous "solution" but I thought I would throw |
5 |
> it out there, and see if any other interested parties (eg. kent\n and |
6 |
> potentially graaff) may be able to share their thoughts.. |
7 |
|
8 |
Actually, the more I think about it, there are 2 major reasons not to do this: |
9 |
|
10 |
1. Its a premature abstraction that centralises a point of failure, and |
11 |
so any defects in this POF scale to all places, producing even more |
12 |
places where changing one eclass requires regenerating the md5 cache |
13 |
for the whole tree, and makes it *harder* to specify logic that only |
14 |
occurs in one languages ecosystem. |
15 |
|
16 |
2. What happens if two eclasses inherit this shared eclass, and |
17 |
declares "which variables do I look into" via other variables that |
18 |
define its API?, and then, one ebuild consumes *both*. |
19 |
|
20 |
Presently, inheriting two eclasses, one for python, one for lua, and |
21 |
taping together how they build inside the ebuild itself seems viable. |
22 |
|
23 |
But if the logic is shoehorned into a common eclass, that will likely |
24 |
go really fucking pear shaped, unless done with a very fucked |
25 |
up/complicated API. |
26 |
|
27 |
I don't think this is worth it. |