1 |
On Fri, Oct 7, 2016 at 7:20 AM, Rich Freeman <rich0@g.o> wrote: |
2 |
> On Fri, Oct 7, 2016 at 10:05 AM, Raymond Jennings |
3 |
> <shentino@×××××.com> wrote: |
4 |
>> My opinion is that if a developer is bad enough to keep out, its |
5 |
>> also |
6 |
>> important enough to get the paperwork fixed to prove it. If they |
7 |
>> have a |
8 |
>> clean case it *should* be very easy to get the paperwork right. |
9 |
> |
10 |
> Sure, by all means leave the bug open until the paperwork is fixed, |
11 |
> but I don't think that means that the developer should be allowed back |
12 |
> in if the bug isn't closed before some deadline. |
13 |
|
14 |
What I want to prevent is a stagnation where a dev gets mistakenly |
15 |
locked out because his case got left in limbo. |
16 |
|
17 |
>> But its the "due process" here that proves the developer is bad to |
18 |
>> begin |
19 |
>> with. If comrel screwed up and there was a mistake and the |
20 |
>> developer is |
21 |
>> actually meritorious, its bad for gentoo to keep them out. |
22 |
|
23 |
> Sure, if all three of your preconditions are true I agree with your |
24 |
> conclusion. However, if comrel screwed up and there was a mistake and |
25 |
> the developer is actually still a problem, then the solution is to fix |
26 |
> the mistakes, not keep them around. |
27 |
|
28 |
And how do you know whether the developer is a problem or not? |
29 |
|
30 |
>> Also, I just realized that the "everyone votes" blocks comrel |
31 |
>> manpower from |
32 |
>> scaling. |
33 |
>> |
34 |
>> What if each member of comrel was granted the discretion to handle |
35 |
>> everything as they themselves saw fit, subject to the provision |
36 |
>> that the |
37 |
>> comrel lead/comrel as a group/the council could receive appeals? |
38 |
> |
39 |
> I tend to agree with this, or a process where Comrel assigns a small |
40 |
> group of devs to each case. The Comrel lead or maybe the entirety of |
41 |
> Comrel could keep a general eye on things as a level of QA and there |
42 |
> would still be Council appeals. I've heard the issue raised that the |
43 |
> system where all of Comrel deals with every case also causes issues |
44 |
> when some members of Comrel aren't around or don't have time to deal |
45 |
> with every case. Basically it makes all of Comrel as fast as its |
46 |
> slowest member. Imagine if every stablereq required every member of |
47 |
> an arch team to approve it? It would be like Ago trying to swim with |
48 |
> concrete shoes. |
49 |
|
50 |
And this is exactly why I suggest breaking comrel up into a team of |
51 |
independent agents overseen by guidelines that preserve uniformity. |
52 |
|
53 |
Being geeks we are all probably familiar with the benefits of |
54 |
parallelization. It's like |
55 |
|
56 |
# make comrel -jN |
57 |
|
58 |
:) |
59 |
|
60 |
And that means comrel shouldn't be drowning in its own red tape. |
61 |
|
62 |
> Indeed, with the model where all Comrel members deciding every case |
63 |
> recruiting more manpower to Comrel would actually have the ironic |
64 |
> effect of lowering their productivity. The best thing you could do in |
65 |
> such a model is instead boot out the slowest contributors. |
66 |
> |
67 |
>>> Imagine if stale stablereqs could be appealed |
68 |
>>> to Council... :) |
69 |
>> |
70 |
>> Maybe not stale stablereqs themselves. This is a volunteer project |
71 |
>> with |
72 |
>> finite manpower. |
73 |
>> |
74 |
>> However, someone deliberately *ignoring* a stablereq when there's a |
75 |
>> large |
76 |
>> need for it MIGHT be more worthy of an escalation. But merely being |
77 |
>> overworked should not be an offense, and it would probably go |
78 |
>> through comrel |
79 |
>> first. |
80 |
> |
81 |
> If you punish people for ignoring important bugs at times, then you |
82 |
> just encourage people to not sign up for the job in the first place. |
83 |
> If you have a single queue multiple server model (which is how most |
84 |
> projects tend to work, including arch teams), then the queue only |
85 |
> stalls if all the servers stall. The solution in this case isn't to |
86 |
> kick out servers, but rather to encourage more to show up and deal |
87 |
> with the high priority bugs (the reality is that our servers don't |
88 |
> strictly work on bugs in priority order, though again I think that |
89 |
> trying to change this could actually lower throughput). Even if a |
90 |
> server only closes one bug per year, that still has a net positive |
91 |
> impact on the queue (assuming zero overhead, which is mostly true for |
92 |
> the way we work). |
93 |
|
94 |
My point is that going after people for deliberately ignoring stablereq |
95 |
is *less* stupid than spamming council with appeals on stale stablereq. |
96 |
|
97 |
I never said it was a good idea, just less bad. |
98 |
|
99 |
The proper solution would probably be an algorithm that flags the most |
100 |
urgent stablereq's. |
101 |
|
102 |
stablereq triage score = age in days * number of reverse deps blocked |
103 |
on the stablereq |
104 |
|
105 |
Google lives and breathes and eats algorithms, and they even use gentoo |
106 |
devs as a recruiting ground (hello recruiters!). |
107 |
> -- |
108 |
> Rich |
109 |
> |