Ross Patterson wrote:
> Rob Miller <robm@...> writes:
> 
>> Cristopher Ewing wrote:
>>> On Jun 19, 2009, at 14:44, Ross Patterson wrote:
>>>
>> - maintaining compatibility across the entire 3.x line of Plone is
>> harder than it should be, more trouble than it's worth.  in this case
>> Plone has broken its contract to maintain 3.x compatibility, and folks
>> should be gently chided on the plone-dev list.  in this case, my
>> suggestion would be that membrane just sucks it up and we live w/ the
>> fact that membrane 1.1 only works w/ Plone 3.0-3.2, and membrane 2.0
>> only works w/ Plone 3.3.
> 
> Can someone summarize exactly what the changes were in 3.3 that so
> egregiously broke things?  Did I miss it in one of the previous posts?

wichert actually knows more about it than anybody.  here's his description 
(originally sent in an off-list email):

<wiggy>
This turned out to be caused by the catalog indexable attributes being lost. 
In Plone 3.3 these are now managed via plone.indexer, which uses named 
adapters internally. For some reason these named adapters are lost in the 
membrane tests. I've modified membrane to use the new API instead, which 
removes a bunch of very verbose deprecation warnings and fixes this problem. 
This does mean membrane trunk now requires Plone 3.3.
</wiggy>

here are the changes that seem to be most related to what he's talking about:

http://dev.plone.org/collective/changeset/87653/Products.membrane/trunk

based on wichert's initial description, it sounds like the older ways of 
registering the indexable object attributes in theory should still work, 
although in P3.3 they've started generating deprecation warnings.  but in 
practice, the indexes were not actually being registered (in the test 
environment, at least), and so he opted to switch to the new API.

-r

-r