1 |
On Sat, 19 Jun 2004 01:27:50 +0200 Carsten Lohrke <carlo@g.o> |
2 |
wrote: |
3 |
| > If an ebuild is missing significant things or conflicting with other |
4 |
| > packages, it should be in package.mask. ~ isn't a dumping ground for |
5 |
| > known broken ebuilds, it's an indication that the package is a |
6 |
| > candidate for stable after testing. |
7 |
| > |
8 |
| As Donnie said: package.mask'ed ebuilds most likely result either in |
9 |
| less bug reports or more users whining about broken stuff, because a |
10 |
| lot of them would unmask nearly everything then. When becoming a dev I |
11 |
| was told, that an ebuild can/should be marked stable thirty days after |
12 |
| the last known bug is fixed. Does every arch team member adhere this |
13 |
| rule and look at bugs.g., if there is an open bug report left and if |
14 |
| the last ebuild change is from a month ago, before declaring an ebuild |
15 |
| stable on xyz? |
16 |
|
17 |
Thirty days? No. What we consider a reasonable period based upon the |
18 |
scope of the package and the degree of the change? Yes. We keep |
19 |
toolchain and base system components ~ for a lot longer than thirty |
20 |
days. On the other hand, a minor bug fix for a non-critical application |
21 |
might go stable after a week. That "30" is a rough guideline, not a hard |
22 |
rule. |
23 |
|
24 |
Surpisingly enough, the arch teams do know what they're doing here. We |
25 |
don't go around marking all kinds of broken packages stable. However, we |
26 |
*will* mark things stable before what Foser calls the "indicator arch" |
27 |
if we feel that it is appropriate. Remember, the core system can be very |
28 |
different version-wise between different archs. |
29 |
|
30 |
What we're doing works. We know from user feedback that stable sparc and |
31 |
mips *are* stable (well, to the extent possible given how b0rked the |
32 |
kernel can be...). Unlike x86, who don't currently have a dedicated |
33 |
team maintaining the stable tree, we don't get people on mailing lists |
34 |
and the forums saying "you might as well run ~arch, it's just as likely |
35 |
to be broken". |
36 |
|
37 |
Case in point: the "indicator arch" for xorg-x11 would be considered |
38 |
x86. If we were to wait for x86 to mark xorg-x11 stable before it went |
39 |
sparc, we would still be stuck with xfree 4.3.0 which (on sparc) doesn't |
40 |
work with 2.6.x kernels, doesn't have working opengl and doesn't have |
41 |
working DRM. As it is, by going ahead of the "indicator arch", we have a |
42 |
far more stable X server which, although it has a few small issues |
43 |
remaining, works a hell of a lot better than the previous version. |
44 |
|
45 |
Or, if you'd like a non-core example... dnsmasq 1.x had strange issues |
46 |
with dhcp on certain sparc boxes, whereas the 2.x series didn't. The |
47 |
same problem didn't appear on x86, so the maintainer was in no |
48 |
particular hurry to switch over 2.x to x86. sparc went ahead of the |
49 |
"indicator arch", resulting in non-broken dhcp for sparc users. Yes, we |
50 |
hit one minor issue because of this, but the more serious problem was |
51 |
fixed as a result. |
52 |
|
53 |
| - From an organizational point of view it would be better, if the arch |
54 |
| teams would file a (P1) request via bug.g.o, giving the maintainer |
55 |
| (herd/s) the possibility to comment it. On the other hand I don't |
56 |
| think Gentoo has enough developers yet, to add another bit of |
57 |
| overhead. :-/ |
58 |
|
59 |
If in doubt, the relevant person is consulted on IRC or by email first. |
60 |
|
61 |
This isn't a case of us consistently going around marking dodgy things |
62 |
as stable. We keyword *as is appropriate*, and the system works. |
63 |
|
64 |
-- |
65 |
Ciaran McCreesh : Gentoo Developer (Sparc, MIPS, Vim, Fluxbox) |
66 |
Mail : ciaranm at gentoo.org |
67 |
Web : http://dev.gentoo.org/~ciaranm |