1 |
On Sat, 2004-02-28 at 21:41, Jason Stubbs wrote: |
2 |
> On Saturday 28 February 2004 12:03, Georgi Georgiev wrote: |
3 |
> > maillog: 28/02/2004-09:22:09(+0900): Jason Stubbs types |
4 |
> > |
5 |
> > > Basically, I'm suggesting that config files that have never been changed |
6 |
> > > be removed as they will be restored upon reinstallation. Anything that |
7 |
> > > has changed shouldn't be removed and anything that was added since |
8 |
> > > installation won't be tracked anyway. |
9 |
> > |
10 |
> > Currently, when a new merge would overwrite a CONFIG_PROTECTed file, it |
11 |
> > creates a file named ._cfg????_originalfile if the file has been changed. |
12 |
> > |
13 |
> > What if something similar is made when unmerging packages? Unmerging a |
14 |
> > CONFIG_PROTECTed file that has not been changed removes the file right |
15 |
> > away. If it was changed, an empty ._cfg????_file is created, with some way |
16 |
> > to distinguish that it is supposed to express deletion (setting the sticky |
17 |
> > bit for example). etc-update can be used to delete the file later, after it |
18 |
> > is updated to recognize empty sticky files as a pointer to delete the |
19 |
> > original file. |
20 |
> |
21 |
> Yep, that's definitely one problem. Another I thought of was when unmerging a |
22 |
> package that provides an init script that has not yet been stopped. Seems |
23 |
> like there are several issues that need to be thought through. Thanks for the |
24 |
> input. |
25 |
> |
26 |
|
27 |
Perhaps ebuilds that provide services should start using the |
28 |
pkg_prerm/pkg_postrm and pkg_preinst/pkg_postinst functions to do things |
29 |
like: |
30 |
Stop the currently running service |
31 |
Start the new service |
32 |
|
33 |
I seem to remember from my days of using RH that when installing a new |
34 |
version of something like apache, once the new version was in place, the |
35 |
service was restarted, but I could be wrong. |
36 |
|
37 |
I know that some people will not like this at all, so perhaps it could |
38 |
be a conditional thing (maybe a variable in make.conf) that defaults to |
39 |
off and has to be explicitly turned on. |
40 |
|
41 |
Or maybe it should only happen if it is an ebuild revision |
42 |
(foo-1.2.3-r1) and not an upgrade, eg: foo-1.2.3 -> foo-1.3.0. |
43 |
|
44 |
-- |
45 |
Troy Dack Gentoo moves pretty fast; if you don't stop and |
46 |
tad@g.o look around once in a while, you could miss out. |
47 |
|
48 |
Public Key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x4D90BE3C |
49 |
Key fingerprint = 1F3D 6C15 16AA 09D5 0C96 92E5 FD89 16F9 4D90 BE3C |