1 |
On 2022-01-09 05:13, gevisz wrote: |
2 |
> Yes, masking some new package can work in this case. |
3 |
> |
4 |
> However, it is not so easy as it may seem because it is not the new |
5 |
> version of tensorflow that I should mask in my case as on the day |
6 |
> when the tensorflow recompilation failed its version remained the same |
7 |
> and only some of its dependencies were supposed to be upgraded. |
8 |
> |
9 |
> Of course, I may try this approach. However, tensorflow is not |
10 |
> considered stable in gentoo tree and it has a lot of dependencies |
11 |
> that are also not considered stable and should be unmasked. |
12 |
> |
13 |
> All this leads to a large number of possible choices on |
14 |
> which packages to mask/unmask. |
15 |
> |
16 |
> So, playing with this is like playing in a casino with about |
17 |
> 4 hours of compilation for each bet. |
18 |
> |
19 |
|
20 |
So you know the date it last compiled and run successfully? |
21 |
|
22 |
If it was me, I'd build a manual list of dependencies (like Dale |
23 |
indicated), then install genlop and run `genlop -t` for each of the |
24 |
dependencies and the main package. It will tell you the versions that |
25 |
were built, and more importantly, the *date* they were built. |
26 |
|
27 |
You should be able to deduce what package versions were working with |
28 |
each other, but then the hard part: trying to figure out if those |
29 |
versions are still available. `eshowkw <package>` will tell you what's |
30 |
available in the tree, but if it isn't available, then it gets way |
31 |
harder as you have to try to find the old ebuilds with sources and |
32 |
possibly set up a local repo and pray those packages don't affect other |
33 |
installed packages. |
34 |
|
35 |
Dan |