1 |
On Tue, Jan 19, 2021 at 09:58:15AM -0500, Matt Turner wrote: |
2 |
> On Tue, Jan 19, 2021 at 6:04 AM Daniel Cordero wrote: |
3 |
> > On Mon, Jan 18, 2021 at 03:21:52PM -0500, Matt Turner wrote: |
4 |
> > > There should be no need for custom updates, and if there are exceptional |
5 |
> > > circumstances the new --enter-chroot option can be used. |
6 |
> > > |
7 |
> > |
8 |
> > There is still a need, and if I'm reading these patches right, |
9 |
> > --enter-chroot would not satisfy that need. |
10 |
> > |
11 |
> > enter-chroot is interactive, and, depending on hardware speed, the time |
12 |
> > between kicking off the catalyst run and when the chroot shell is |
13 |
> > invoked may be some time for a human operator to monitor. |
14 |
> > |
15 |
> > It also adds non-deterministic, un-reproducible behaviour to a build. |
16 |
> > Having it in the spec file at least means it is included in a build |
17 |
> > artifact. |
18 |
> |
19 |
> Can you think of a case where update_seed would be needed? I tried, |
20 |
> and couldn't think of anything, but I might not be imaginative enough. |
21 |
> |
22 |
> My thinking is that with @changed-subslots, there's essentially no |
23 |
> need to update the seed stage further. But there have been some |
24 |
> exceptional cases that don't really fit update_seed (e.g., things that |
25 |
> cannot be fixed with emerge ...). |
26 |
|
27 |
update_seed has been very useful when dealing with perl/python version |
28 |
upgrades, where large swathes of packages need to be updated at the same |
29 |
time. I'm not sure if @changed-subslots has in any way resolved this, |
30 |
but it's convenient to have an override. |
31 |
|
32 |
Another example was a recent bash/readline update, where the stage3 had |
33 |
readline.so.5 and built the new stage1 bash against that, but installed |
34 |
readline.so.6 into the stage1root, hence the stage2 build was broken. |