Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Fw: [platform-update-dev] Feature changes for 3.0


Injected a couple of comments below. Overall it sounds ok though.


Thanks,

-----------------------------------
Peter Manahan
WebSphere Tools
Build/Install and
Product Architecture
------------------------------------
manahan@xxxxxxxxxx

----- Forwarded by Peter Manahan/Toronto/IBM on 11/11/2003 10:05 AM -----

[platform-update-dev] (no subject)




° From: Dorian Birsan <birsan@xxxxxxxxxx>
° Date: Tue, 4 Nov 2003 13:39:01 -0500
° Delivered-to: platform-update-dev@xxxxxxxxxxx



We have been doing some simplification of the update/install component, mostly in the UI and some code refactoring for supporting command driven operations, and now we're investigating the update API's, including the xml specs.
One item that seems to play a major role in the increased complexity (code-wise and the number of bugs associated with it) is the <include> feature relationship.  The include relationship has been introduced to allow features to be built out of existing features, as opposed to having to always list all the plugins needed.  The final plugin list is obtained by transitive closure.
However, in Eclipse 2.1 <include> was no longer a strict include, it allowed more flexibility by using the "match" attribute:
http://dev.eclipse.org/viewcvs/index.cgi/%7Echeckout%7E/platform-update-home/doc/working/feature_extensions_2.html
One main reason was to allow updating of nested features independently of parents, and another reason was to allow feature patching.
We would like to go back to the previous direct include (no "match" attribute) and address the two reason differently:
- treat <include> as ownership and do not allow nested features to be updated independently of the parent features

<pm>
that was the original intention I believe. The policy items mentioned below would be required to help ensure there were not updates/fixes applied by parties not
owning the features. We need to be able to prevent updates to WSAD for example from Eclipse.org for support reasons.
</pm>


- patches will no longer include features, but only plugins.  A patch will be associated with a root feature, and contain plugins from any nested feature.  A patch should also be able to <include> other patches.

<pm>
Although in theory I consider this to be a bad thing
   <rant>
         Plugins should have one and only one owner/residence/delivery mechanism because once you move into the range of hundreds of  plugins shipping in a varity of products
              it becomes important that they are controlled from one central point. Else there are mismatches and errors and confusion ensures. Generally those who argue against it only have to worry about
              a small set of plugins.
    </rant>

This is acceptable for patches I guess as long as the policy below is implemented.
</pm>

- if you have a need for a nested feature needs to be updated independently of its parent, then don't use <include> for it, but make it an imported feature, with the appropriate match attribute.

<pm>
        I am not sure what use case this would be for? In general we'd  like to have the whole product included under one umbrella feature. Currently we can't because the
          include option doesn't work across sites and we have three sites. So the user sees 3 updates for the product when in reality there is only one update eg 5.0->5.1). This bullet seems to indicate that
        that would increase the number of updates visible to the user when in fact we need them to decrease.  Unless this is the general case and the problem of cross-site includes is being fixed :-)
</pm>

- in addition, we are playing with the idea of a policy file that determines what can and can't be updated, and from what update sites.

<pm>
     This seems to me to be required in order to keep the current function. I assume it would play a part when searching for updates as well?
So if you find a site with an update that isn't in the policy file you ignore it.  Addtionally it would be nice if there were a way to add
</pm>
 
Please post any comments about the proposed changes and, if they are breaking changes, please describe your scenarios.
Thanks,
-Dorian

Back to the top