| 1 |
On Thu, Mar 26, 2009 at 12:15, Caleb Cushing <xenoterracide@×××××.com> wrote: |
| 2 |
> On Wed, Mar 25, 2009 at 10:11 PM, Jonas Bernoulli <jonas@×××××××.li> wrote: |
| 3 |
>> |
| 4 |
>> For various reasons "etc-under-version-control" loses much of the |
| 5 |
>> power and/or ease of use that comes with the underlying vcs. Much of |
| 6 |
>> this being due to the fact that the files being kept in the repository |
| 7 |
>> are totally unrelated. |
| 8 |
>> |
| 9 |
>> The solution to this obviously is to maintain a separate branch for |
| 10 |
>> each package. |
| 11 |
> |
| 12 |
> I have to ask, what's the 'obviousness' here? |
| 13 |
|
| 14 |
What I basically said is "the problem is that unrelated files live in |
| 15 |
the same branch, the solution obviously is to use separate branches". |
| 16 |
While one might disagree that this actually is a problem the |
| 17 |
conclusion can't really get much more obvious - it might even qualify |
| 18 |
as a tautology :-) |
| 19 |
|
| 20 |
> I don't see why each |
| 21 |
> package should have it's own branch, if they are not related, |
| 22 |
|
| 23 |
Because this allows you to say "these changes I have made to $package |
| 24 |
are grap, make them go way" (git reset --hard HEAD~n). Without |
| 25 |
seperate branches you would have to create a new commit in which only |
| 26 |
some files where reverted to previous state (git checkout HEAD~n -- |
| 27 |
file; git commit file). This clutters history. |
| 28 |
|
| 29 |
Or some user has problem configuring a given program, and asks on the |
| 30 |
forum. With separate branches she could be pointed at a particular |
| 31 |
branch in some repository which contains a working configuration for |
| 32 |
that application. She certainly is not interested in the config files |
| 33 |
of some other applications she has never heard about before. |
| 34 |
|
| 35 |
> A they should have their own repo, |
| 36 |
> B, they should be in the same branch. |
| 37 |
|
| 38 |
They should be in different repositories but in the same branch? That |
| 39 |
ain't gonna work :-) What do you mean by this? |
| 40 |
|
| 41 |
(I think you are saying that each package should have it's own |
| 42 |
repository. Some things you said however seam to contradict this.) |
| 43 |
|
| 44 |
> their relation is relative as they are all configs for a system. |
| 45 |
|
| 46 |
True. But not all config files are related. Setting up a screensaver |
| 47 |
is totally unrelated to your firewall. |
| 48 |
|
| 49 |
Some things are related however and a particular configuration of |
| 50 |
package A might only make sense when package B is configured a certain |
| 51 |
way. Therefor confsink has to support something like @sets. |
| 52 |
|
| 53 |
> I'm a big git user so I'm trying to understand conceptually what you're on about. |
| 54 |
|
| 55 |
Think git-submodule where the submodules do not have to live in |
| 56 |
dedicated subdirectories. I will elaborate on this shortly. |
| 57 |
|
| 58 |
This is way having all branches of all packages in the same repository |
| 59 |
is better than having them split into separate repositories. You can |
| 60 |
easily add some other file tracking the states of the installed |
| 61 |
"submodules". Of course it is also possible to track this information |
| 62 |
in some other way: in another branch in an other branch, but this |
| 63 |
complicates things unnecessarily. |
| 64 |
|
| 65 |
Actually implementing git-intermixed as a less restrictive alternative |
| 66 |
to git-submodules would be the very first step towards confsink. |
| 67 |
. |
| 68 |
-- Jonas |
Archives