Moinak Ghosh wrote:
> On Sat, Jun 7, 2008 at 1:46 AM, Bart Smaalders <[EMAIL PROTECTED]> wrote:
>> Brock Pytlik wrote:
>>> Bart Smaalders wrote:
>>>> Michal Pryc wrote:
>>>>
>>>>> Hi,
>>>>> I have already asked about this few months back:
>>>>> http://mail.opensolaris.org/pipermail/pkg-discuss/2008-March/002217.html
>>>>>
>>>>> There is an script which shows how the description are taken from the
>>>>> manifest file. To get GUI up and running we need to have descriptions
>>>>> for *all* packages, so currently this is quite long task, more then that
>>>>> this is even longer if the manifests are not in the local catalog.
>>>>>
> [...]>>>>
>>>>> ./print_changes.d /path/to/the/ips/catalog
>>>>>
>>>> Rather than caching, etc, I'd like to suggest that using the search
>>>> mechanism to retrieve information is a better  approach.
>>>>
>>>> This would allow for the quick retrieval of descriptions, icons, etc,
>>>> from current or new repos.
>>>>
>>>> - Bart
>>>>
>>>>
>>> I'm not sure how you're envisioning using the search mechanism to do
>>> this. To my knowledge, search doesn't currently support getting the
>>> description (for example) of a specific package. This wasn't a use case
>>> for search which I had thought of so I'll have to consider whether the
>>> new version of search would have the capacity to do this (at least if it
>>> provides (or could provide) the data structures to make this efficient)
>>>
>>> Also, setting aside whether or not search could do this, isn't the cost
>>> of transferring all of this information from a remote server on start up
>>> each time prohibitive? Especially for things like icons, isn't that a
>>> large amount of data to pull across the network given the time frame (<
>>> 1 sec) we're considering?
>>>
>>> If I understand correctly, there are really two problems: pulling
>>> information for all the local individual manifest files takes too long;
>>> pulling the information for manifests not stored locally takes way too long.
>>>
>>> On the first, I suppose there's no reason (other than duplication of
>>> data) not to gather this information and pkg install time and add it to
>>> a central location. On the second, given that hitting the server seems
>>> unavoidable to me (which means the 1sec constraint is likely to be
>>> violated), perhaps what should be done is the program caches a small
>>> subset of what it will need to retrieve from the server (which would
>>> probably be gui dependent since different gui's might prioritize
>>> different packages) then it can show the user the cached information
>>> while it goes off and asks the server for updated information on those
>>> packages and fills in the rest of its package descriptions.
>>>
>>
>> A small local cache is fine as a gui hack, but I'd much rather work towards
>> a model of retrieving descriptions and icon pointers dynamically.  The icons
>> can be cached locally by hash.
> 
>    It is fine to fetch information as and when the user clicks on the package
>    name. However this constant dependence on network connection does
>    not sit well with laptop mobility. User with laptops on the move will be
>    unable to check package information half the time. One use case I can
>    think of is that a user has to do a bunch of package operations and he
>    meticulously makes package selections/deselections while on a plane/
>    train and stores this as batch operation to be processed later. Later
>    when he reaches and gets onto a network he just fires off the batch
>    operation and does other work. This or similar scenarios won't be
>    possible if pkg metadata is not locally cached.
> 

Are there any extant systems where users actually do this?  It seems a 
little too theoretical to me at first glance.  It would also seem to 
raise a host of issues around the effects an out-of-sync manifest cache 
would have on such a stored batch operation.  All in all, I'm skeptical 
I'd want it.

Dave
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to