On 03/20/09 09:00, Alok Aggarwal wrote:
> Hi Jan,
>
> I'm just now catching up on this discussion ..
>
> On Fri, 20 Mar 2009, jan damborsky wrote:
>
>> The current plan is to behave in the same way as if old AI image
>> was used - that means if empty packages list was provided,
>> default list would be picked up from ai_manifest.defval.xml.
>>
>>> I understand we want to support the old tag, but why not
>>> support the old tag, but require at least 1 package name be specified?
>>
>> The reason is that current default AI manifest doesn't contain
>> list of packages, so it wouldn't be functional with new AI image.
>>
>> That said, may be we try to be too much nice in this point.
>> Given the fact that AI is evolving pretty quickly and more
>> fundamental changes are likely to occur, this might not be
>> the on the current list of rules it is reasonable to follow.
>> I am open to suggestions :-)
>
> I like this plan. I do however question the need for
> having an ai_manifest.defval.xml at all at this point.
>
> The only reason it was there was to meet the requirements
> of the XML defaults validator. Since in the case of AI,
> it really doesn't convey any useful information that can't
> be captured in the ai_manifest.xml itself, but rather
> obscures the workings of AI, I think we should just try
> getting rid of ai_manifest.defval.xml altogether.
>
> The default authority/repo as well as default values
> for partitioning/slicing can just be migrated over to
> ai_manifest.xml. The AI observability would be much
> better in that case.
>
> What do you think?
Hmmm... the file does seem a bit sparse. Getting rid of the defaults
the file specifies and being explicit in ai_manifest.xml does, as you
say, make everything more observable. On the flip side, sometimes
having a program assume defaults makes it easier for the user to setup.
In this case there are only a few defaults set though, so removing them
won't make usability that much easier. The benefit of defaults use
increases with the number and complexity of the manifest.
Before getting rid of the file, consider another question: can the
defaults and validation mechanism be put to better use in AI? I suggest
it can. Rather than hardcoding lots of sanity checks in AI itself, you
an leverage the validation mechanism to do such checks as "are any
packages specified at all" very easily. But that is another issue for
another time. You can leave the file in place for now, but with nothing
between the <defaults_and_validation_manifest> tags.
Thanks,
Jack
>
> Alok
> _______________________________________________
> caiman-discuss mailing list
> caiman-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
