Tricks for laying down good foundations ;-)

[Rahul Sundaram]

Ignacio Vazquez-Abrams wrote:
> On Mon, 2008-12-15 at 18:03 +0530, Rahul Sundaram wrote:
>> Seth Vidal wrote:
>>> On Mon, 15 Dec 2008, David Timms wrote:
>>>> Jesse Keating wrote:
>>>>> On Mon, 2008-12-15 at 07:54 +1100, David Timms wrote:
>>>>>> Currently, there appears to be no defined order for install process; 
>>>>>> in fact of a default-ish install of 1000 packages, kernel might be 
>>>>>> around 850 to 900 into the process, openoffice around 500.
>>>>> The order is determined by rpm itself, after sorting out all the install
>>>>> loops, the %pre/%post requirements, etc...
>>>> Let's say that yum asked to install certain packages, or groups that 
>>>> met the above criterion, as 4x separate transactions. Then rpm would 
>>>> have to obey wouldn't it ?
>>>>
>>> Sure, but what does that have to do with anything?
>> Obviously he is looking at a problem and trying to provide ideas to 
>> improve the situation. I assume, the fundamental idea of installing the 
>> base dependencies first if that makes sense should be a enhancement 
>> request to rpm instead of working around it in either Anaconda or yum.
> 
> How far down the rabbit hole do we go? Certainly if coreutils is
> important, then why not shadow-utils? sed? dos2unix? Do we abolish
> Requires(pre) altogether?

I guess something like, if a package that is in @base or @core, is part 
of the transaction set, install it first, would work?  The installer, 
could work in two stages. The first stage gets you basic bootstrappable 
system. This is a small number of packages + boot loader write, so a 
failure due to issues like a bad burn is minimized. So even if the 
subsequence stage fails (bad media, missing packages etc), you still get 
something you can work with and a restart could possibly resume the 
installation as well.

Currently, failures can and often does leave you with a system that 
doesn't boot. Not very ideal.

Rahul