1 |
On Tue, May 31, 2011 at 12:05:03AM +0200, Andreas K. Huettel wrote: |
2 |
> Am Montag 30 Mai 2011, 23:55:52 schrieb Brian Harring: |
3 |
> > If someone has a definition that is commonsense, then propose it- the |
4 |
> > current "you must log everything" is very, very heavy handed and |
5 |
> > basically was a forced situation since QA cannot make folks behave |
6 |
> > when the rules are reliant on common sense. |
7 |
> |
8 |
> Well how about "any change that can influence the behaviour of (portage|your |
9 |
> favourite package manager) in any way or present the user with different |
10 |
> output"? |
11 |
|
12 |
Mostly correct, although the problem there is 'influence'; consider: |
13 |
|
14 |
src_unpack() { |
15 |
bunch of code |
16 |
} |
17 |
|
18 |
being changed into |
19 |
|
20 |
new_func() { |
21 |
bunch of code |
22 |
} |
23 |
src_unpack() { |
24 |
invoke new_func |
25 |
} |
26 |
|
27 |
That should have no influence, thus not being ChangeLog'd. The |
28 |
problem is when the dev screws up, and it *has* an influence but no |
29 |
ChangeLog. |
30 |
|
31 |
A more real world example is people abusing eval and other things |
32 |
(python eclass for example)- folks can/do argue that it has no |
33 |
influence, but the complexity means it may have unexpected affect. |
34 |
|
35 |
That's the crux of the issue, and it comes down to common sense. |
36 |
Come up w/ a sane policy for things like that and I'll owe you some |
37 |
beer. |
38 |
|
39 |
Either way, for the rest of it, as Diego said, LGTM. I'm just |
40 |
nitpicking here to make it absolutely clear to people where we start |
41 |
running into policy issues. |
42 |
|
43 |
~brian |