invalidateAllObjects can break other EC's that are currently editing objects 
because you throw away the backing snapshots from underneath them ...

ms

On Feb 11, 2010, at 8:45 AM, Jean-Francois Veillette wrote:

> I'm surprised no one mentionned ec.invalidateAllObjects() coupled with 
> refetching arrays you might have in memory (displaygroup and such for 
> example), as far as I understand, it's the only solution that will take care 
> of relationship and get rid of deleted objects (faults will still be in 
> memory but at least refetching relationship and arrays will avoid pointing to 
> them).
> 
> - jfv
> 
> Le 10-02-08 à 14:20, Mike Schrag a écrit :
> 
>> yeah, this is why i'm suspicious that we'll see a generalized Wonder 
>> implementation of this .... definitely some tricks we could do, like what 
>> you're saying -- just changing attributes that don't participate in 
>> relationships, inverse relationships, or restricting qualifiers could be a 
>> relatively easy update. changing anything that participates in a 
>> relationship would be sort of a pain -- you have to do that pre-fetch thing 
>> first and then you'd have to fake notifications afterwards. for delete, it's 
>> even nastier.
>> 
>> i think you take advantage of the knowledge you have of your special case 
>> and custom write this. it's topics like these that make me sometimes think 
>> the everything-is-cached approach is overkill. i'd love to see a variant of 
>> EOF that lets you write like a stateless framework in cases where you don't 
>> want all the snapshotting stuff.
>> 
>> ms
>> 
>> On Feb 8, 2010, at 2:13 PM, Anjo Krank wrote:
>> 
>>> Mostly, it depends on what you are doing. Changing, say, status=done is 
>>> different from owner=<people pk:1>, because the one only changes internal 
>>> state, the other touches relationships.
>>> 
>>> Then again, all your *other* ECs in all *other* instances won't get 
>>> notified anyway (unless you use the ERCNF). So your code needs to be able 
>>> to handle that problem anyway.
>>> 
>>> Cheers, Anjo
>>> 
>>> 
>>> 
>>> Am 08.02.2010 um 20:02 schrieb Mike Schrag:
>>> 
>>>>> Mike's precautionary measure is ticking at the top of my mind... so may 
>>>>> be for the time being I will just call ec.refreshAllObjects() just to be 
>>>>> integral, consistent, simple and more importantly let not annoy EOF by 
>>>>> mistake!!!
>>>> my precautionary tale is about using the methods you're using at all (i.e. 
>>>> the updateRowsDescribedByQualifier) ... you're sneaking behind EOF and 
>>>> basically doing direct DB operations. you're then trying to come back and 
>>>> expect an easy way for EOF's caches to be in-sync with your changes. the 
>>>> general case here is that you can't do it without tossing all your 
>>>> snapshots, because you have no idea if the snapshots in your cache are 
>>>> actually in-sync with the current state of the database when you executed 
>>>> your update. there's a reason EOF does what it does when you perform all 
>>>> of these operations, and it's because it actually needs to.
>>>> 
>>>> probably the closest-to-right way to do this is to prefetch the rows that 
>>>> would be updated or deleted, perform the operations, then use the GIDs to 
>>>> ... i guess manually do everything that EOF would have done. you're going 
>>>> to lose all the inverse relationship updating, and you're going to lose 
>>>> delete rules, etc. also, by fetching into the EC beforehand, you're 
>>>> basically taking the performance hit that you were probably trying to 
>>>> avoid in the first place by using those API's.
>>>> 
>>>> so i doubt there's a simple generalized API that will go into Wonder for 
>>>> this -- i'm not people would be happy with the performance profile of it.
>>>> 
>>>> ms
> 


 _______________________________________________
Do not post admin requests to the list. They will be ignored.
Webobjects-dev mailing list      ([email protected])
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/webobjects-dev/archive%40mail-archive.com

This email sent to [email protected]

Reply via email to