1 |
On Wed, 2022-02-09 at 00:54 -0600, William Hubbs wrote: |
2 |
> On Wed, Feb 09, 2022 at 09:23:46AM +0500, Anna Vyalkova wrote: |
3 |
> > On 2022-02-09 04:15, Sam James wrote: |
4 |
> > > A compromise / workaround would be the Go maintainers working on |
5 |
> > > some tooling to make it easier to tar up the needed modules and |
6 |
> > > avoid huge SRC_URI. |
7 |
> > |
8 |
> > This. Sometimes EGO_SUM is so long it exceeds the maximum string length. |
9 |
> > |
10 |
> > *blames google for the whole go modules thing* |
11 |
> |
12 |
> This isn't an issue specific to Go; Rust can have the same problem. |
13 |
> |
14 |
> I'm speaking mostly for Go; I'm not sure how Rust handles this exactly. |
15 |
> |
16 |
> If we could define an extra phase function that runs after |
17 |
> src_unpack, say call it src_fetch_extra, which has access to the |
18 |
> network, we could completely eliminate EGO_SUM and most of the go-module |
19 |
> eclass code around src_unpack. |
20 |
> |
21 |
> At a very high level, if src_fetch_extra is not defined, nothing |
22 |
> happens. If it is defined, however, it would be called with "${S}" as |
23 |
> the starting directory. |
24 |
> |
25 |
> Could this be added to eapi 9? |
26 |
> |
27 |
> If I had something like this for Go ebuilds at least, all of the big |
28 |
> manifests would disappear. |
29 |
> |
30 |
|
31 |
This would: |
32 |
|
33 |
1) break Gentoo installations behind restrictive firewalls, |
34 |
|
35 |
2) make parallel fetching much harder, |
36 |
|
37 |
3) would require custom implementations to support caching |
38 |
and mirroring, |
39 |
|
40 |
4) will eventually lead to ebuilds fetching and using unverified data. |
41 |
|
42 |
-- |
43 |
Best regards, |
44 |
Michał Górny |