1 |
Dan Armak posted <200410192040.54826.danarmak@g.o>, excerpted |
2 |
below, on Tue, 19 Oct 2004 20:40:54 +0200: |
3 |
|
4 |
> On Tuesday 19 October 2004 20:23, Roman Gaufman wrote: |
5 |
>> err, DO_NOT_COMPILE="kmail" in /etc/make.conf does just that, and |
6 |
>> supports every package or even service KDE has to offer. Feel free to |
7 |
>> not compile ksmserver or kinit as well -- this will break things, but |
8 |
>> the fact is, gentooists always had the possibility to do this! |
9 |
>> |
10 |
>> What exactly do you propose here? -- do you actually propose gentoo |
11 |
>> developers should split the metapackages into close to 100 ebuilds? -- |
12 |
> |
13 |
> I'm not proposing anything. I'm announcing the fact that I and another |
14 |
> Gentoo developer (motaboy) have actually split the monolithic ebuilds |
15 |
> into well over 300 new packages. |
16 |
|
17 |
Wow! |
18 |
|
19 |
>> what gain over DO_NOT_COMPILE does this give? |
20 |
> |
21 |
> It allows portage to manage interdependencies and, in fact, everything |
22 |
> else it manages. Heavy use of DO_NOT_COMPILE, such as emerging only |
23 |
> kmail + its deps from all of kdepim (of course each user has to figure |
24 |
> out for himself what those deps are, first), is more or less equivalent |
25 |
> to linux from scratch. It'd be manual building, except portage will |
26 |
> think it knows what is installed, and will be wrong. |
27 |
|
28 |
Interestingly, I just bit the bullet with 3.3.1 and configured all those |
29 |
DO_NOT_COMPILEs for myself. Some packages (kdebase, kdegames) I emerged |
30 |
intact, others (kdeadmin, kdepim) I killed more than half the package. |
31 |
kdepim was the worst to try to handle, as I had to backtrack on a couple |
32 |
items (libical and libkcal) and compile them anyway due to dependency |
33 |
issues. KMail and KNode were the main apps I wanted out of it, plus stuff |
34 |
like the kfile metadata helpers. kdeartwork I just killed the |
35 |
screensavers. A couple packages (kdeedu) I don't install at all anyway. |
36 |
|
37 |
I had originally tried to use KDE_REMOVE_DIR from the ebuilds, but found |
38 |
some stuff that didn't need actually compiled but DID need the dirs still |
39 |
there for headers, etc. (kpilot was one of these, altho it wasn't |
40 |
compiled anyway due to use flag.) I went back to DO_NOT_COMPILE. |
41 |
|
42 |
Good point re portage tracking. How do you manage the dependencies? Do |
43 |
you rebuild every time in-builddir, or do you look in the deployed system |
44 |
first and only build in-builddir if it isn't yet deployed? Is there a |
45 |
list of an optimal emerge order if the latter? |
46 |
|
47 |
I've been thinking about confcache. I haven't tried using it yet, but I'm |
48 |
certainly thinking about it. Does it just do the main stuff, or does it |
49 |
handle the entire configure step? From earlier, I remember you (?) saying |
50 |
what it DID cache, it invalidated entirely if there was just one change. I |
51 |
can see how this would be simpler to implement. However, if you are |
52 |
caching everything, I'd hope you are doing it in segments, so that for |
53 |
instance the normal C compile checks on endianness and bitlength and the |
54 |
like, the "main" checks, are not invalidated when something like qt |
55 |
changes. Perhaps section it into "main system qualities" like endianness |
56 |
and the like that aren't likely to change, "standard bintools" |
57 |
checks,"gcc", "kde/qt", "gtk/gnome", and "misc" (off the top of my head)? |
58 |
That would maintain a grouping, but wouldn't invalidate the entire |
59 |
confcache for one change to the kde dependencies with the 300 kde packages |
60 |
you mention. As I said, however, I haven't yet used it, just been turning |
61 |
the idea over in my head, so... |
62 |
|
63 |
-- |
64 |
Duncan - List replies preferred. No HTML msgs. |
65 |
"They that can give up essential liberty to obtain a little |
66 |
temporary safety, deserve neither liberty nor safety." -- |
67 |
Benjamin Franklin |
68 |
|
69 |
|
70 |
|
71 |
-- |
72 |
gentoo-dev@g.o mailing list |