1 |
Hello, everyone. |
2 |
|
3 |
There has been some unrest wrt Python 2 removal lately. To put things |
4 |
back in order, I've spent a considerable time filing bugs for all |
5 |
the remaining packages yesterday (some of them turned out to be false |
6 |
positives, sorry) and I'd like to issue some official guidelines |
7 |
to avoid misunderstandings. |
8 |
|
9 |
|
10 |
The official deadline for Python 2 in Gentoo is 2021-01-01. This means |
11 |
that up to this date, all packages have to be ported to Python 3, with |
12 |
new versions stable and the last old versions ready to be cleaned up. |
13 |
All unported packages will be last rited with 30 day removal period |
14 |
and no possible extension. |
15 |
|
16 |
I would really like to avoid extending the deadline. That is, unless |
17 |
the big G. or the big M. forces us to. Apparently they're too busy |
18 |
inventing new programming languages to port their scripts. |
19 |
|
20 |
However, we shouldn't wait till the last minute. It makes sense |
21 |
to wait for packages that are high-profile and/or have good chances |
22 |
of actually being ported. It doesn't make sense to delay removing some |
23 |
obscure package that hasn't seen any activity in 10 years. The whole |
24 |
point is that we last rite packages in smaller batches that make it |
25 |
easier to deal with, and that we last rite them early to give more time |
26 |
for users to react. It would really suck if we did a 250-package last |
27 |
rites on January 1st. |
28 |
|
29 |
|
30 |
Now, to the guidelines: |
31 |
|
32 |
1. Normally, last rites apply only to packages that *unconditionally* |
33 |
need Python. Please *do not* last rite packages if it is sufficient |
34 |
to mask USE=python or alike. |
35 |
|
36 |
1a. Masking or removing conditional Python support is not need |
37 |
immediately necessary; however, it may be desirable to do it early |
38 |
to give users more time to complain, and to unblock dependencies |
39 |
if there are any. |
40 |
|
41 |
1b. A corner case here are packages requiring Python for tests. |
42 |
Restricting tests suck. |
43 |
|
44 |
1c. In case of some unmaintained packages, where other treecleaning |
45 |
conditions apply, it may be desirable to last rite anyway. |
46 |
|
47 |
|
48 |
2. Give people a chance to react. Last rite only if there is no reply |
49 |
to the relevant bug for, say, 30 days. |
50 |
|
51 |
2a. If maintainers point out that the work is underway, it's fine |
52 |
to wait. |
53 |
|
54 |
2b. But if the work has stopped progressing for a long time now |
55 |
and it is unlikely that it will be completed before the deadline, |
56 |
it may be better to stop waiting and warn the users. |
57 |
|
58 |
|
59 |
3. Do not last rite reverse dependencies without warning. If you |
60 |
notice some last rite candidate is blocked, please *at the very least* |
61 |
CC maintainers of revdeps to give them notice that their packages are |
62 |
affected. Filing extra bugs is even better. |
63 |
|
64 |
3a. I suppose extra bugs shouldn't be necessary if revdeps belong to |
65 |
the same family of packages and have the same maintainer (i.e. you can |
66 |
clearly guess they need to go as well). |
67 |
|
68 |
|
69 |
4. Leave high-profile and/or likely-to-be-ported packages for later. |
70 |
At least right now we have enough obscure and dead packages to last |
71 |
rite first. |
72 |
|
73 |
4a. That doesn't mean wait with them forever. It would be nice not to |
74 |
last rite all high-profile packages in one go. |
75 |
|
76 |
|
77 |
5. Please *verify* everything before proceeding. As I've mentioned, |
78 |
some bugs might be false positives. Some packages may be fixed without |
79 |
closing the bug. |
80 |
|
81 |
5a. Also note that some packages have || deps that show up in rdep |
82 |
reports but don't justify last riting reverse dependencies. |
83 |
|
84 |
5b. Please file stabilization bugs (but do not CC arches without |
85 |
approval) or ping maintainers wrt cleanup whenever appropriate. |
86 |
|
87 |
-- |
88 |
Best regards, |
89 |
Michał Górny |