1 |
On Sun, 2004-05-02 at 20:56 -0700, Greg KH wrote: |
2 |
> On Sun, May 02, 2004 at 10:04:35PM -0400, Nathaniel McCallum wrote: |
3 |
> > The new glep is here: http://www.gentoo.org/proj/en/glep/glep-0026.html |
4 |
> |
5 |
> My formal objections: |
6 |
> |
7 |
> Basically the idea is nice, but I think the proposed solution is not |
8 |
> correct. Here is why I feel this way: |
9 |
> |
10 |
> > Stage 1 would be achieved through a seperate utility (perhaps like the |
11 |
> > current genkernel). This utility would help the user configure the |
12 |
> > kernel and any kernel/build related settings. This stage is optional |
13 |
> > and would only be used if a person wanted a customized kernel |
14 |
> > settings. One optional thought is to have this utility a part of the |
15 |
> > base Gentoo system. That way there are less steps to make a custom |
16 |
> > kernel. |
17 |
> |
18 |
> Why reinvent the wheel here? The kernel has a _wonderful_ configuration |
19 |
> system already. It's provided in about 4 different interfaces at least |
20 |
> (command line, ncurses, gtk, qt), and in library form. Are you |
21 |
> proposing hooking into that library to do the configuration? If so, |
22 |
> what is lacking in the current implementation that is needing another |
23 |
> tool? |
24 |
|
25 |
ok, perhaps the name is a little misleading. The program itself doesn't |
26 |
generate .config files, it just manages all aspects of creating them. |
27 |
To actually create the file, it just calls "make menuconfig" or whatever |
28 |
is appropriate. However, what it does do is manage the config files |
29 |
(not create them). And by that I simply mean making sure they get |
30 |
copied to the right place, etc... This is not really much more than |
31 |
genkernel does (if at all) |
32 |
|
33 |
> > Stage 2 would be implimented through an eclass. This stage is not |
34 |
> > optional. One would perform this step by typing "emerge kernel-name", |
35 |
> > where "kernel-name" is the name of the kernel package you want to use |
36 |
> > (ie. "gentoo-dev"). This package would have a "buildkernel" USE flag. |
37 |
> > If the flag was not set, it would simply download and install sources |
38 |
> > like we do currently. However, if the "buildkernel" flag is set, |
39 |
> > portage will perform the following steps: |
40 |
> > |
41 |
> > 1. (as a dependency) download and install a tarball of "generic" |
42 |
> > kernel config files. |
43 |
> |
44 |
> Is this really needed? Don't we already do this today? |
45 |
Yes, but: |
46 |
A. It isn't its own package. Therefore, kernel configs cannot be |
47 |
version tracked by users. this is not good open development practice. |
48 |
B. The BUILD system isn't integrated into portage. |
49 |
|
50 |
> Does this work? As specified before, I really think you need the whole |
51 |
> tree around in order to build external modules. Now with 2.6, building |
52 |
> those modules is a world easier (1 line makefile), and it works if you |
53 |
> are not root (no write access necessary). But again, all of this is |
54 |
> already being addressed in the genkernel rewrite that is happening (and |
55 |
> you can see it already in cvs.) |
56 |
|
57 |
I'm pretty sure you don't need them all. However, even if you do, with |
58 |
this idea, you can just keep them tarballed, rather than unpacked. This |
59 |
still saves space. |
60 |
|
61 |
> |
62 |
> > Stage 3 would merely be an ebuild that constructs an initrd image for |
63 |
> > either the running kernel or, lacking a running kernel, the newest |
64 |
> > kernel installed (by version/date installed?). Initrd's can't be used |
65 |
> > on all arches, so this ebuild can be keyword masked as appropriate. |
66 |
> > The initrd package would also have a "bootsplash" USE flag (on x86, |
67 |
> > maybe others) that would build in bootsplash support. Any non-default |
68 |
> > actions desired by the user can be handled with the utility from Stage |
69 |
> > 1. |
70 |
> |
71 |
> So "bootsplash" will be the only USE flag? Why stop there? |
72 |
> |
73 |
> Oh, and initrd images are dead. I predict that by the end of this year |
74 |
> we will not be using them at all, as there is no need to (other distros |
75 |
> are already proving this is possible.) We also get a much more |
76 |
> flexible, portable (works on all platforms) and cleaner implementation |
77 |
> with initramfs, so that is what we will be moving to soon. So your |
78 |
> "initrd" options are pretty moot. |
79 |
|
80 |
Fine, but people use them now, so we need to support them. However, |
81 |
with the abstraction of initrd into its own ebuild, when initrd's go by |
82 |
the wayside, just make a new initramfs ebuild. |
83 |
|
84 |
> |
85 |
> > There are many advantages gained by this method: |
86 |
> > |
87 |
> > 1. Full arch support (GentooInstaller really could use this) |
88 |
> |
89 |
> We should fix genkernel to do this, if it doesn't already today. There |
90 |
> is no reason this should not be possible today without a total rework. |
91 |
|
92 |
True, but why fix something that isn't integrated into portage? Plus, |
93 |
it is much easier to fix/maintain things when they are cleanly |
94 |
abstracted. This is not at all the case with genkernel. |
95 |
|
96 |
> > 2. More consistent with the rest of portage (installs binaries |
97 |
> > by building from source) |
98 |
> |
99 |
> True, it isn't as "clean" |
100 |
> |
101 |
> > 3. Better user experience |
102 |
> |
103 |
> You forgot "mom and apple pie" here too :) |
104 |
|
105 |
You joke, but think about being able to manage kernels from porthole |
106 |
(http://porthole.sf.net)? or another application? It really does add to |
107 |
better user experience. |
108 |
|
109 |
> > 4. More flexability (genkernel forces building of initrd on x86) |
110 |
> |
111 |
> I've already outlined the issues of initrd going away, so this isn't an |
112 |
> issue. |
113 |
|
114 |
Even if initrd DOES go away, all the more reason to integrate into |
115 |
portage. I mean, you'll be rewriting over half of genkernel anyway! |
116 |
|
117 |
Better to just abstract the stuff first... |
118 |
|
119 |
> > 5. Don't have to store full kernel sources (each source tree is |
120 |
> > ~200MB unpacked) |
121 |
> |
122 |
> You don't have to that today if you don't want to build any other kernel |
123 |
> modules that are outside of the main kernel tree. I have a number of |
124 |
> boxes set up this way right now. In fact, those boxes do not even have |
125 |
> a single kernel package on them at all, I just rolled my own and |
126 |
> installed it. Once of the niceties of Gentoo is that you do not have to |
127 |
> install a Gentoo kernel in order to have a working box (which for my |
128 |
> job, is a very good thing.) |
129 |
|
130 |
But try installing a kernel, building it, then emerge -C'ing it. You |
131 |
are left with all the extra junk behind. YES, I know that this is the |
132 |
correct behavior, and I agree. However, it is sloppy. Adding the |
133 |
building into portage cleans that up while still maintaining correct |
134 |
behavior. |
135 |
|
136 |
|
137 |
> > 6. Creates the ability to make binary kernel packages for fast |
138 |
> > installs (think GRP kernel) |
139 |
> |
140 |
> I see no where in your proposal for how this will be accomplished. In |
141 |
> fact, this is the first place in the GLEP that "binary kernel packages" |
142 |
> are even mentioned. Please either add the description of how you are |
143 |
> going to do this to the SPECIFICATION portion of the document, or take |
144 |
> it out of the RATIONAL portion. |
145 |
|
146 |
I would think this is rather self evident. To over-simplify, when you |
147 |
use the "-b" option with emerge it makes a tarball of all the files you |
148 |
install and appends the ebuild to the end of the file. If building is |
149 |
handled by portage, and it is "installing" not source files but binary |
150 |
files, the "-b" option will make a tarball of the kernel binary, the |
151 |
modules, and the kernel-headers/sources. all you have to do to install |
152 |
this kernel now is an "emerge -k kernel". This is the same exact |
153 |
fashion as all other packages. And thus, I don't think it needs |
154 |
clarification. |
155 |
|
156 |
> |
157 |
> > 7. A seperate package for "generic" kernel config files gives us |
158 |
> > better version tracking |
159 |
> |
160 |
> "better version tracking"? What do you mean here? Kernel config files |
161 |
> can not be kept identical across 2 different kernel versions. And they |
162 |
> are not supposed to be. They are a vital portion of the kernel version |
163 |
> itself and can not be split out. To think otherwise is to not fully |
164 |
> understand the kernel build process. |
165 |
> |
166 |
> Also, is this a problem today? I have not seen issues related to this |
167 |
> in the past, have you? Are there any bugs you can point us at that |
168 |
> would show how this would help out? |
169 |
|
170 |
OK, the way genkernel currently does this is by storing a "template" |
171 |
config file for each major kernel version (ie. 2.4, 2.6, etc). When |
172 |
genkernel decides to use the "default config", it copies the "template" |
173 |
config to the source directory and does a "yes | make oldconfig". I'm |
174 |
not proposing anything different than this. I'm MOSTLY talking about a |
175 |
restructuring of genkernel, NOT a start from scratch. The restructuring |
176 |
abstracts the config from the kernel building from the initrd/initramfs. |
177 |
This is cleaner, and in the process, gives us portage integration. |
178 |
|
179 |
By version tracking I simply mean this: I have a tarball of "generic" |
180 |
kernel configs. Lets call it "sys-kernel/generic-kernel-configs" for |
181 |
fun. The first time I release them it is version 1.0. If I find a bug |
182 |
and fix it, we now have version 1.1. This lets our users know when |
183 |
something has changed in the kernel config. Its just more "open" and |
184 |
that is really all I meant by it. |
185 |
|
186 |
At this point I'd also like to say that this GLEP is a proposal. It is |
187 |
in "draft" stage. So by all means, propose ideas. But don't just say |
188 |
"what we are doing now is great" because, even though it is, it could be |
189 |
better. That is all I'm pointing out. |
190 |
|
191 |
Nathaniel |
192 |
|
193 |
|
194 |
-- |
195 |
gentoo-dev@g.o mailing list |