This makes it possible to "reload" the Ptrs in active cells when changes happen to the ESM record that they are based on. In practice, the old Ptrs are deleted, their RefNums and MpNums are blanked out, and new Ptrs are created that use the same RefNum and MpNum as before.
The above has required me to also add a method called setRefNum() to CellRef to allow setting a RefNum on the fly.
There may be a more elegant implementation available for updatePtrsWithRefIds(), but it requires additional research.
Previously, attempts to reuse the same reference for multiple creature disguises led to movement animation issues, as well as a dynamic_cast error in Creature::getInventoryStore() that made a DedicatedPlayer vanish completely when they first lost their creature disguise, then disguised themselves as a creature that could not hold weapons and then disguised themselves as a creature that could hold weapons.
The usage of const_cast has been replaced with usage of MWWorld::getModifiableStore() and ESMStore::overrideRecord()
Methods whose names started with "update" now start with "override", for consistency with ESMStore's overrideRecord()
New methods have been added for "overriding" enchantment, potion and spell records, which actually leads to them being created with their already set refIds if they haven't been created yet, as per the description of ESMStore::overrideRecord(): "Insert a record with set ID, and allow it to override a pre-existing static record."
Usage of RecordHelper methods has been updated in DedicatedPlayer.
Actors who are on the ground have their inertial force ignored, so they are now made to not be regarded as being on the ground in World::setInertialForce()
I originally added rotation animation sync as part of commit 068a45be87. Unfortunately, it meant the PlayerPosition packets were now twice as large as they had been before, which was less than ideal for such a frequently sent packet, which is why Koncord switched to a more optimized approach in commits 5f30dfd5db and d67db1a9bd.
Recently, there have since been some rotation animation problems in OpenMW, which have broken the way Koncord's approach looks. My original approach still looks somewhat okay, so I'm switching back to it until we can figure out how to reuse it under the current circumstances.
Although weather sync was added by Koncord to the rewrite in fd721143e2 in a way that used surprisingly few lines of code, it relied on the server requesting weather states every second from authority players and sending them to non-authority players, while also allowing only very sudden weather transitions across regions, i.e. if there was one player in the Ascadian Isles who had stormy weather, and another player with clear weather in the Bitter Coast Region walked across to the Ascadian Isles, that player was instantly made to have stormy weather with no kind of transition at all.
My approach solves both of those problems. It solves the packet spam by only sending weather updates to the server when weather changes happen or when there are new arrivals to a weather authority's region, and it allows for both sudden weather transitions when players teleport to a region and for soft, gradual transitions when players walk across to a region. It is inspired by my previous actor sync, and uses a WorldRegionAuthority packet to set players as region authorities in a similar way to how ActorAuthority sets players as cell AI authorities. Weather changes are created only by the region authority for a given region, and weather packets are also only sent by that authority.
However, it should be noted that gradual weather transitions are used by default in this implementation. To use sudden weather transitions, the serverside Lua scripts need to forward WorldWeather packets with the forceWeather boolean set to true. That is, however, already handled by our default Lua scripts in situations where it makes sense.
This should clarify the real meaning of the packet and its associated event.
The event itself has been renamed from OnPlayerKillCount to OnWorldKillCount.
The ActorPacket-sending functions now have sendToOtherVisitors and skipAttachedPlayer arguments, except for the ones for ActorList and ActorAuthority (because such arguments don't make sense for those).
The last received ActorList can now be copied into the write-only ActorList that can be sent in packets. Changing the pid of the write-only ActorList can now be done separately from clearing its contents.
The server can now make actors activate players and objects, at least in theory. In practice, OpenMW''s AiActivate package needs to be worked so it allows specific objects as targets instead of just refIds.
Unfortunately, disabling clientside deletion of summons and items that can be picked up requires extra work on actors and inventories respectively, to avoid buggy situations.
Unfortunately, default values set in the C++ code for our script function parameters don't actually seem to work, and they always default to false because they receive a nil value from Lua. As a result, to not break compatibility with previous scripts, I've decided to use a skipAttachedPlayer argument instead so it can default to false while still providing the same benefits that sendToAttachedPlayer provided.
Previously, there was a confusing separation between script functions that had a "broadcast" argument and script functions that had a "toOthers" argument.
Those with broadcast sent the packet to all players on the server when broadcast was true. Those with toOthers sent the packet to all players other than the packet's attached player.
The former was based on the pattern of the original SendMessage() script function. The latter more closely resembled RakNet's own broadcast argument as seen here:
https://github.com/TES3MP/CrabNet/blob/master/include/raknet/RakPeer.h#L219
This commit makes it so all sending functions have a sendToOtherPlayers argument that is false by default and a sendToAttachedPlayer that is true by default. This should simultaneously allow sending to be more intuitive, while not breaking previous existing scripts to a significant degree.
Additionally, this commit also reduces some code repetition for all instances of packet-fetching in script functions.
Previously, all crime witnesses stopped being hostile to a respawning player for as long as the player's diedSinceArrestAttempt was true. That meant that, in an area with no guards to arrest the player, crime witnesses did not enage in combat with the player at all ever again until diedSinceArrestAttempt became false.
This commit makes it so the time of the last crime is recorded for each witness, and that is then compared with the time of the LocalPlayer's last death for a one-time crime forgiveness during that player's current life.
This is essentially a gameplay adjustment for "singleplayer with respawns," and will have to be reworked to make sense for every player in multiplayer, though that requires reworking the crime system as a whole and is thus on hold.
Add serverside script functions for determining the killers of both players and actors.
Use unsigned ints for script functions returning an object or actor's refNumIndex or mpNum.
Remove updateDeadState() from LocalPlayer and make its code part of updateStatsDynamic() for simplicity.
This prevents summon duplication caused by a SummonKey sometimes being deleted immediately after being created, before the server can send back an ObjectSpawn packet spawning a creature that can be attached to the SummonKey.
Capostrophic
Andrei Kortunov
Marc Zinnschlag
David Cernat
Koncord
Bret Curtis
Capostrophic
terrabyte25
Alexander Stillich
Doc West