[arch-general] Kernel Module Package Guidelines

[Michael Towers]

Aaron Griffin wrote:
> On Dec 2, 2007 6:19 PM, Robert Emil Berge <list at rebi.no> wrote:
>   
>> On Sun, 2 Dec 2007 15:23:57 -0600
>> "Aaron Griffin" <aaronmgriffin at gmail.com> wrote:
>>
>>     
>>> I still can't understand why this is a problem though... according to
>>> pacman, the installed size is 12K, and the only possible reason I
>>> could think of for caring about this dep is size.... could you please
>>> explain the rationale here?
>>>       
Although that wasn't my main point, the rationale in the case of aufs is 
that the utils package modifies the behaviour of the mount command by 
providing the (normally unnecessary) helper script mount.aufs, which in 
turn depends on various other packages (sed, grep, awk, diff, ...). It's 
no big deal, and these will normally be installed anyway, and it is also 
possible to stop the helper script from being called, and ... It's just 
that in the case of aufs - which might well be used in a situation where 
these utilities aren't available - it could be a bit annoying, 
especially if the aufs-utils dependencies were fixed to include all 
these things which I might not want (and which would increase the total 
size to significantly more than 12k).

>> I understand Michael's point, I think. He's not talking about the aufs
>> package at all, but using it as an example for the rule in the Packaging
>> Guidlines that says modules should always depend on their utilities,
>> even when you can use the modules without them. He wasn't complaining
>> about not getting things exaclty as he wants them, he was only asking a
>> curious question. He wants to know the reason for this exception to the
>> rule of packages only depending on what the package needs to be useful.
>>     
Exactly.
>> To me it seems your answer is: We don't have a reason, and stop
>> bothering us with stupid questions. Or is it; it's ok with deps that
>> are not necessary as long as they're small?
>>
>> Although it's a bit pedantic, I think he has a point too. If you should
>> follow this principle all the way, the kernel26 package should depend on
>> cryptsetup, nfs-utils, dosfsutils, fuse, iptables, ntfsprogs etc., you
>> get my idea..
>>     
>
> The rationale is that the aufs and aufs-utils packages are actually
> part of the exact same source tarball and are simply separated due to
> the fact that we support multiple kernels. As such, the _original
> author_ intended them to be used together.
>
>
>   
squashfs-tools and unionfs-utils are also part of the same source 
packages as their respective modules, but the kernel doesn't depend on 
these. In the case of aufs, the original author says in the 
documentation that the utils are optional (even if he did intend them to 
be used together with the module, he didn't necessarily intend the 
module to be used together with the utils).

I can live with the situation as it is, I actually only wanted to point 
out a possible inconsistency and an easy and painless way to remove it.