patch is attached that replaces the o(n^2) algorithm currently
used to build the inventory cache with an o(n) algorithm using
hash tables. the patch also adds some additional error handling.
* This is a HUGE OMG update and will definitely have unknown side effects.. so this is really only for the strong hearted at this point. Regular people should let the dust settle.
* This has been tested to work with most basic functions. However.. make sure you back up 'everything' before using this. It's that big!
* Essentially we're back at square 1 in the testing phase.. so lets identify things that broke.
occasionally, but are fixed on restart (issue 1919).
This patch introduces the following changes:
1. when a user teleports out of Region A, remove that user's profile
from the Region A user profile cache
2. when a user crosses between regions out of Region A, remove that
user's profile from the Region A user profile cache
3. the user profile cache's session ID member can now be set (written),
and is updated each time a connection with a new avatar is established (ie: a
new avatar enters the region)
4. when a region server looks up a user profile and a cache miss
occurs, fetch the user profile from the user server first instead of
immediately returning null
* May resolves inventory problems that occur when the user moves between two regions`
* e.g. if the user moves to a second region, adds an inventory item, moves back to the original region then tries to manipulate that item
* Not yet implemented for teleport
And reverted CachedUserInfo back to revision 5262.
I don't think most of the inventory problems that people are reporting are due to these, but its easier to deal with one set of potential problems at a time, and I'm not going to get any time in the week to work on this anyway.
TODO next is to make the login server read/write a users inventory from the correct server (the inventory url set in a userprofile)
On the region side, although not tested with multiple servers it should work if that inventory url was set, and the inventory servers urls have been added to the CommunicationsManager, using CommunicationsManager.AddInventoryService(string hostUrl)
Fixes:
- Wearable icon and name sreset to default on copy/paste
- Cache is not updated when renaming/moving folders
- Partial refactor to make inventory less dependen on AssetBase having a "Name" field
- Add llGiveInventoryList() function
* Detaching from right clicking in world, detaches to your inventory.
* If you go up to a prim and attach it from in world, it appears in your inventory.
* Attachment placement is saved when you detach them.
* Choosing wear remembers your last attachment point from inventory.
* Wrote a method to update an inventory item's asset and sends the updated inventory item to the Client
* Wrote a recursive method to find the folder of a known existing inventory item.
* Removed a block on physics object position on creation. This might crash a region or two, let us know via Mantis if your region crashes because of a physics out of bounds error.
* Drop doesn't work. The menu item doesn't even come up. Don't know why :P.
* This replaces the old behaviour of failing straight away, which could cause lost updates if the inventory service was slow in responding
* This is the first baby step to making all inventory requests behave this way, to reduce inventory lossage
* This allows caps requests to be routed to regions where the agent is currently a root agent instead of the region that they logged into as it did previously.
* This fixes a wide variety of bugs related to 'can't do X once i've crossed a border'.
* The first seed cap request fails, the second one works. (this generates an error message on the console)
* Experimental.
* A much larger race condition where the inventory request could be made before the region gets any data at all will also be fixed in the near future.
* This change also fixes a regression from two patches ago where items stopped appearing in standalone inventory
