Patches from Second Life Mobile (August 2024) #95
Conversation
* GridManager requests now properly handle case sensitivity (i.e. all region lookups should be case invariant) * Adds bidirectional lookups to GridManager allowing handle lookups as well as name. Caches.
…m. Better to start small, then increase later; so we can prioritise what we get first; otherwise it'll be too late, and we're already getting tons of far away objects we didn't want.
…. If you're living in the same data center as the simulators, this can go faster.
…t isn't implemented. * Logout reply timer can now be cancelled, such as if it is successful. * OnSimDisconnected event now has a "Was Connected" property, this'll be false if OnDisconnected is firing before the simulator connected in the first place. * Adds a reason for disconnection events in logs * Prevents an exception if a session is cancelled before it is fully connected with the _packet?box objects. * Adds a FindSimulator overload which takes a handle rather than just an IPEndPoint * Region handshake sends 0x7 instead of 0x2 in flags. This impacts the order and speed of object loading.
…nated between various running apps. Time since startup is not a useful bit of information when coordinating with simulator and client logs.
…orrect. This is because it only uses 32-bits, the official viewer uses 64-bits, and the overflow matters. * Adds a .Valid property to TextureEntryFace - sometimes it is not, and will throw exceptions if you actually try use anything in this class. At least have the ability to check first.
… agent is currently on.
* I'm not a big fan of this code to be perfectly honest. It probably can be fixed better. There's possibly an error on the simulator side, or in this code, that I haven't managed to locate, when the EventQueue has an early return rather than long-polling as it should, and it's triggered by this code. Not sure if the official viewer suffers the same fault (possibly.)
… scale, locking so frequently actually results in a nasty performance issue. * EventQueue be checked periodically to make sure it's running. Sometimes it dies. This is a blunt fix (along with the previous commit), but is needed to ensure that event queue-y things keep working. * Adds a 500ms sleep between connecting to a simulator and CompleteAgentMovement - there's a race condition within the simulator which sometimes results in a faulty login, this seems to help. * SetSeedCaps has an extra parameter, if the user's simulator hasn't changed, don't do anything. * Add a lookup which lets you turn full Object UUIDs into LocalID, while LibreMetaverse prefers localID when referencing primitives, the official viewer uses UUIDs - and several things actually work better this way. (For example, avatars can spontaneously change LocalID sometimes). This way we can follow those changes.
…ueue is down. * Adds some functions to check if a message is a normal agent message, or a group message. * Adds new RelativePositionEstimate to complement RelativePosition - RelativePosition actually ends up drifting badly. RelativePositionEstimate instead calculates each time it is requested, using known velocity, and is far more accurate in practice. * Adds an override to ChatEventArgs.ToString to dump information about the chat. * Improves reliability of Group Messages * Teleports previously used GetGridRegion which is sometimes unreliable, request the full map, once, if it fails. The results will be cached. * Adds AgentManager.LastPositionUpdate to tell you when the avatars position was last updated
…me in 12 years. Preserves the OpenSim specific packets and additions.
* DEFAULT_AGENT_UPDATE_INTERVAL is now configurable * INTERPOLATION_INTERVAL is now configurable * You can now disable periodic SEND_AGENT_UPDATES independently (SEND_AGENT_UPDATES_REGULARLY) of normal updates, such as in CompleteAgentMovement * You can disable TexturePipeline if you don't need it * Fixes a possible timing bug in AgentManagerMovement.TurnToward
…atar.Attachments - this needs a Simulator change to support it, as it is a new message property in Avatar messages. * This can be used for determining if an avatar is "completely loaded" from the network side, along with the next commit...
* Adds the ability to check how many children a primitive has, and should have. * Adds Primitive.GetChildren which returns all known children of a primitive. * Known Gotcha: ChildCount will count the self as well. * If ChildCount != Primitive.GetChildren(...).Count()+1 -- then the linkset is incomplete, and you may want to do a properties request on it to get sent the missing children.
…end your own AvatarID instead of the target AvatarID * Adds ChildCount property to AvatarManager for determining the number of attachments that should be present * Improves reliability of AvatarAppearance messages
* Correctly deserializes legacy materials
* Adds new methods for requesting legacy materials * Completes switchover of ObjectPrimitives from InternalDictionary to ConcurrentDictionary * For clients using OBJECT_TRACKING, this may be a performance regression in busy deployments. * Adds CRC support to object updates, and adds this to tracking * Changes isNewObject to isProbablyNew to more accurately describe the property * Improves InterpolationTimer methods (but we don't use it - so please test this if you do.)
…ges to the displayed group title: * Clear cached name, groupName in Avatar when an ObjectUpdate has new NameValues. * Some message paths didn't maintain group title, while others did. (Although I don't think that affects this implementation.) * In the proxy, do send an avatar update for group title changing to empty string. * ActivateGroup and RequestAgentDataUpdate commands from UI to app to proxy to sim. * Message back to UI OnAgentDataUpdate, so that we can tell whether a given group of this agent is already active.
…taUpdateMessage messages * Fixes XML comments
LibreMetaverse/AgentManager.cs
Outdated
| Client.Network.SendPacket(move, simulator); | ||
| Logger.Log("Sending complete agent movement to " + simulator.Handle + " / " + simulator.Name, Helpers.LogLevel.Info, Client); |
There was a problem hiding this comment.
Prefer string interpolation in future commits.
There was a problem hiding this comment.
Agreed, Resharper usually hits me over the head about that one.
There was a problem hiding this comment.
Count it fixed. ;)
|
Just had a review of the Codacy report, it's a bit pedantic, eh? I have fixed a few of the things it has pointed out, but I'll argue against the remainder of its suggestions.
|
|
|
||
| if (responseString.Contains("502 Proxy Error")) | ||
| { | ||
| // LL's "go ask again" message. |
There was a problem hiding this comment.
Oof, always wondered. Assumed it was a squid error.
There was a problem hiding this comment.
Yeah, it's weird. I asked one the Simulator team about it, and they said it's haunted, and they'd love to replace it with something better.
There was a problem hiding this comment.
OMV crashes on it. Used to be a big issue because of all the bots constantly crashing and logging back in.
There was a problem hiding this comment.
Yeah, the issue is actually a little deeper. The problem is after this happens, it no longer "long polls" the HTTP request. The server team doesn't want an instant retry, so we kind of switch to 'active polling' here afterwards (which is my biggest issue with this code), which'll poll again in seconds. One of the subtle bugs we fixed in this version is to keep incrementing the message counter when this happens, and not sending the "isDone"=true, which keeps it somewhat working when it reaches this state.
We've added one more thing here which is the .RestartIfDead() method, this way if you're expecting an important message, like for example, teleports, or such, you can at least force an instant 'check for new messages'; rather than wait for the next natural reconnect.
This is almost certainly a Simulator bug, but as I said, code's haunted. (The Simulator dev might not have used those exact words, I think they actually said 'It's old, flaky and unreliable; and we want to replace it.')
|
Argh, one moment. That was probably me when fixing the merge errors. |
|
Sorry, I did something really dumb, and edited our internal fork, instead of the external repo. This is going to take me about 15-30 minutes to fix. (The sooner we can drop the internal one, the better...) |
|
No worries. I have a prior engagement I need to get to so I will review when I get back. |
* Renames _HttpCts to _httpCancellationTokenSource * Disposes of _httpCancellationTokenSource in finaliser
* Implements IEquatable<Attachment> on Attachment
…GroupName to better explain their purpose. * Removes redundant assignment
….com/secondlife/libremetaverse into secondlife-mobile-patches-august2024 # Conflicts: # LibreMetaverse/Capabilities/EventQueueClient.cs
|
Okay, so - it's compiling now. So that's a win. TestClient needs some further updating, which I am doing now. I'm a little bit more scared now - mainly by f394c71 (the sync with HEAD) which appears to have done quite a lot I didn't want it to. Going to do a bit more review before I'm happy with this - I'll signal when to restart the review. |
|
Okay, I THINK it's okay now. We'll do some testing on our side with this version, and just verify that we've not functionally broken anything. |
|
Sorry, spoke too soon -- still got some issues. This might take me into early next week to iron out. I'll ping when we're ready for another retry! Enjoy the weekend! |
This is entirely just my opinion, but.... The people that do complain about a breaking changes are almost always just trying to offload the maintenance and improvement work they should be doing on their products onto the package maintainers. I see it as exploitative and no package maintainer should be obligated to take on the extra work of maintaining backwards compatibility cruft onto themselves. So unless there is some kind of paid support contract to maintain something, It is almost always better for the package to break stuff. |
|
Many, not all, changes have been cherry-picked into master branch. Didn't make the ObjectPrimatives change thus far. Still considering the impact. It would be advisable to sync the changes before continuing. Thank you! |
|
Oh brilliant thankyou! Sorry we went a little quiet, I was actually getting our team to do the same thing -- break each change up so merging would have been easier. In terms of the signature change; we're about to do the work to change the local indexes to uuid -- maybe it makes the most sense to simply wait for that, and we'll close this PR. |
|
Agreed. Looking forward to the progress! |
This contains a large collection of fixes, improvements and changes from the Second Life Mobile project to date. It's from a forked internal version which has been focusing on the features needed by SL Mobile - so; avatars, appearance, inventory and so on.
Each commit has notes on what has changed, in addition, each commit (unfortunately) contains more than one change; so you'll want to see the full commit notes for each on what was changed.
Some notable highlights:
_Packets_.csto match the latest message template (I noticed OpenSim had a few custom additions relating to VarRegions and such, so we've kept those, and patched them back manually.), this hadn't been done in a long, long time - so there's quite a lot of small changes in here. Testing notes: OpenSim might want to check these changes are still okay upstream.Full Change List
ChildCount != Primitive.GetChildren(...).Count()+1-- then the linkset is incomplete, and you may want to do a properties request on it to get sent the missing children. Returns -1 if the data is unknown.