Gentoo Archives: gentoo-dev

From: Rich Freeman <rich0@g.o>
To: gentoo-dev <gentoo-dev@l.g.o>
Subject: Re: [gentoo-dev] Changing order of default virtual/udev provider
Date: Tue, 09 Feb 2016 12:39:17
Message-Id: CAGfcS_mzHqWxHVMr4H3937TfUVV--CrPnT0UfnRKGmF6qdbV-w@mail.gmail.com
In Reply to: Re: [gentoo-dev] Changing order of default virtual/udev provider by "Anthony G. Basile"
1 On Tue, Feb 9, 2016 at 7:14 AM, Anthony G. Basile <blueness@g.o> wrote:
2 > On 2/9/16 6:59 AM, Rich Freeman wrote:
3 >> On Tue, Feb 9, 2016 at 12:27 AM, Anthony G. Basile <blueness@g.o> wrote:
4 >>> On 2/8/16 10:09 PM, Rich Freeman wrote:
5 >>>> How many of those 14 distros have more than 14 users?
6 >>>
7 >>> gentoo is very unpopular as a distro. however, it excels as a meta
8 >>> distro. if you marginalize its special features, you take away all its
9 >>> charm.
10 >>
11 >> Gentoo's special feature is that it is source-based, not that it uses
12 >> a different udev implementation from everybody else by default.
13 >
14 > that's not what i said.
15
16 You implied that I'm marginalizing Gentoo's special features. I don't
17 see how using a different udev implementation than 99% of the linux
18 install base is one of Gentoo's special features.
19
20 But, by all means clarify what you do mean if I'm misunderstanding
21 your point. Or, if it
22
23 >
24 >>
25 >>>>
26 >>>> Look, I get it, some people don't like systemd. That's fine.
27 >>>> However, you have to realize at this point that a non-systemd
28 >>>> configuration is anything but mainstream.
29 >>>
30 >>> neither is a system based on musl or uclibc, but if you need an embedded
31 >>> system, then these are "mainstream".
32 >>
33 >> Sure, but they're also not defaults.
34 >
35 > this is circular argumentation. they are not default where they are not
36 > default. and they are where they are.
37 >
38 > you simply want to privilege a certain set of targeted systems
39 > (desktop/server) over another set (embedded).
40
41 Gentoo's default virtual providers are almost universally focused on
42 desktop/server systems.
43
44 Even if you accept the argument that eudev is more suitable for
45 embedded, it would be just one more virtual in a sea of virtuals whose
46 defaults are not ideal in the embedded world.
47
48 If somebody wants to create an extension to profiles where you can
49 change the order of virtual defaults so that you can have an embedded
50 default that preferes uclibc over glibc more power to them.
51
52 However, in distro we're maintaining today everything is focused on
53 desktop/server.
54
55 >
56 >>
57 >>>
58 >>> anyhow, the argument in the subject is the order of udev and eudev in
59 >>> the virtual, not systemd vs eudev.
60 >>
61 >> And that is about defaults.
62 >
63 > nope. currently stages come with sys-fs/udev as default.
64
65 The whole point of this thread is that somebody wants to change this.
66 Inevitably changing the defaults requires having a big slug-fest over
67 which new default is better. And that is why I made the suggestion to
68 try to moot the decision by just making this part of the handbook.
69
70 Of course this is going to be a contentious discussion. Of course
71 people are going to get upset about this.
72
73 I'm sorry if my suggestion to make the default less relevant seems
74 like a diversion to the original point of the thread, but I actually
75 see that as the only way to actually RESOLVE the thread. Otherwise
76 this just turns into a which-implementation-is-better slugfest, which
77 is what we're having right now.
78
79 >>>
80 >>> There will always be a
81 >>>> "poppyseed linux" whose purpose in life seems to be to preserve linux
82 >>>> without sysfs or some other obscure practice.
83 >>>
84 >>> no, more like special uses. you're framing the issue based on your
85 >>> notion of "mainstream"
86 >>
87 >> My notion of mainstream, and Fedora's, and Debian's, and RHEL's, and Arch's...
88 >
89 > correct, that is your notion. in the grand scheme of things windows is
90 > mainstream and fedora, debian, etc are just marginal.
91 >
92 > you're missing the point that the style of argumentation you're taking
93 > is one of a particular view that you privilege.
94
95 No argument that different people use Gentoo for different reasons,
96 and they're going to weigh more heavily the arguments that emphasize
97 their own preferences.
98
99 The problem is that ultimately with the design of Gentoo there can
100 only be ONE default virtual/udev provider.
101
102 There are many ways to resolve this:
103 1. Somebody just unilaterally changes the default. (Not likely to
104 work - just leads to revert wars and nonsense in the repo.)
105 2. Eventually the council is asked to pick a winner. (The council
106 will likely not want to do this, and no matter what they pick a huge
107 portion of the distro will probably be upset about it).
108 3. Make the default unimportant by making this part of the handbook.
109 (This is what we do with emacs vs vim, and it has the side benefit of
110 being more appropriate for installs that don't even need a udev
111 implementation.)
112
113 Historically Gentoo has tended to take the approach of #3. This
114 avoids having to wage huge battles over stuff like this. I think that
115 would be more productive than the two of us going back and forth about
116 whose baby is uglier.
117
118 >
119 >>
120 >>>>
121 >>>>>
122 >>>>> it needs to be in the new stage4s to make a bootable system. imo a
123 >>>>> stage4 should be bootable modulo a kernel.
124 >>>>>
125 >>>>
126 >>>> Sure, a stage4 based on systemd makes a lot of sense.
127 >>>
128 >>> not for embedded and the things i work on. these have users.
129 >>>
130 >>
131 >> Systemd makes plenty of sense for many embedded solutions. For the
132 >> kinds of solutions where it doesn't make sense, I'm not sure that
133 >> linux makes sense.
134 >
135 > like? embedded os has avoided systemd like the plague. they opt for
136 > busybox mdev.
137
138 For a box where RAM is at a super-premium, even the Linux kernel may
139 be too much to ask for.
140
141 If RAM isn't at much of a premium (hundreds of MB), then systemd runs
142 just fine and it has a lot of features that support homeostatis which
143 I'd think would be an important feature in an embedded system. Its
144 only real downside is its newness.
145
146 >
147 > the assumption here is that Gentoo is a single distribution where its
148 > from source nature makes it a meta distribution. so you create a false
149 > dichotomy between gentoo and "tiney distros whose main purpose" Many of
150 > those "tiny distros" are gentoo because gentoo is a set covering many
151 > derivatives.
152
153 And those distros can still exist no matter what the default udev
154 provider is. I'm all for preserving choice. I just don't think that
155 eudev is an appropriate default for the sorts of use cases where any
156 of our other default providers make sense.
157
158 >
159 > the reason i'm engaging in this is not because of the default. the
160 > reason i'm engaging in this dialogue is because of the repeated
161 > reduction in vision as to what "from source" means.
162 >
163
164 Again, I'm not suggesting taking away choice at all. In fact, my
165 suggestion was to move this into the handbook to give users more
166 choice than they have today.
167
168 There is already a thread on gentoo-user asking how to safely switch
169 from udev->eudev. If that were just a part of the handbook that isn't
170 even a migration they'd have to make on a new install. Nor would
171 udev->systemd.
172
173 To summarize my goals in this thread:
174 1. Suggest that rather than picking a winner we instead just let the
175 user pick the winner.
176 2. Demonstrate the futility of actually trying to pick a winner,
177 because we all have different values on this subject.
178
179 --
180 Rich

Replies