1 |
On 26/10/19 23:35, William Hubbs wrote: |
2 |
> On Sat, Oct 26, 2019 at 11:17:18AM +0200, Michał Górny wrote: |
3 |
>> On Sat, 2019-10-26 at 11:14 +0200, Dirkjan Ochtman wrote: |
4 |
>>> On Sat, Oct 26, 2019, 05:59 Kent Fredric <kentnl@g.o> wrote: |
5 |
>>> |
6 |
>>>> On Fri, 25 Oct 2019 15:03:39 -0700 |
7 |
>>>> Georgy Yakovlev <gyakovlev@g.o> wrote: |
8 |
>>>> |
9 |
>>>>> not used anymore |
10 |
>>>>> |
11 |
>>>>> Closes: https://bugs.gentoo.org/695698 |
12 |
>>>>> Signed-off-by: Georgy Yakovlev <gyakovlev@g.o> |
13 |
>>>> Its likely this removal will cause the same kinds of problems faced by |
14 |
>>>> the recent virtual/pam removal, just its more insidious, as the |
15 |
>>>> dependency on the virtual is hidden away inside an eclass. |
16 |
>>>> |
17 |
>>>> But this still means that anything users have already installed will |
18 |
>>>> still depend on this, and without --changed-deps=y, it will break |
19 |
>>>> portage's resolution of anything currently installed using this crate. |
20 |
>>>> |
21 |
>>>> You can work-around this by -r1 bumping everything that used this |
22 |
>>>> eclass .... but this just goes to show why there's policy against |
23 |
>>>> eclasses changing the dependencies of their consumers without any |
24 |
>>>> consumer involvement. |
25 |
>>>> |
26 |
>>> In most if not all cases, this is just a build-time dependency. Do we |
27 |
>>> really have all these problems for build-time only dependencies? |
28 |
>>> |
29 |
>> Yes. Because of --with-bdeps. |
30 |
> I disagree, build-time dependencies can change in place because they |
31 |
> only affect the build. The problem with virtual/pam was that it was a |
32 |
> runtime dependency as well. |
33 |
> |
34 |
> William |
35 |
The problem is that portage defaults to --with-bdeps=y, so any emerging of |
36 |
packages now triggers anything that has a build-dep change, unless, as |
37 |
previously stated, you exclude that case or change the defaults. |