If there is a problem reading the blacklist file from disk, previously,
null would be returned from the DDSaveHandler.readBlacklist method. This
would result in a crash later on down the line when we tried to iterate
the blacklist. Now we return an empty array. I'd like to be able to
diagnose the issues causing the blacklist file to be corrupted, but
unfortunately when this issue crops up, people only post the crash :(
If the server disconnected you due to a timeout or whatever, the pocket
manager would appropriately unload all dimension & pocket data, but it
would do so moments before the world unloaded, meaning that entities would
sometimes, on update or as part of their renderer, attempt to access the
dimensional data. This places a new guard condition that attempts to stop
that from crashing the client.
On startup, existing sub-dimensions of personal dimensions were being
given the wrong worldprovider by the launch code. The reason for this is
that subdimensions are not added to the player -> dimension mapping for
personal dimensions, because there is only one canonical "personal"
dimension. The launch code was using presence or absence in the mapping
to determine whether a dimension is personal, rather than the dimension
type. This is now fixed.
Some users were reporting very sporatic crashes in
generateLightBrightnessTable() when the server was starting up. It's
possible that very occasionally, the server will register the pocket
dimension before any pocket dimension data is initialized, forcing the
light brightness table to get generated server side with bad values.
In some cases (such as another mod crashing on startup or something), the
server shutdown even was getting called before anything was actually
initialize. In these cases, some uninitialized objects were having
cleanup methods called on them, when they were null. There are now guard
conditions to prevent the resulting NPE's.
Major change is addition of fractal rift rendering, currently first
pass.
Curves are registered and pregenerated in mod_pocketDim.
Rifts look up these curves, choose one, rotate it, and render it.
The render is a TESR that does stuff. Hard to explain, look at
RenderRift in the code and look at the actual rifts in game to get an
idea of what it does.
I had to add a triangulation library to accomplish this. Will hopefully
do something else that drag around all this.
(I tried(and used comments))
Fixed a crash from manipulating rift data on the client side. I let this
happen because it seemed like TileEntityRift already did that before.
This crash also exposed another issue: that server-side functions are
being used on the client side. I'm not sure how pervasive this is but
some client dimensions are being constructed with the server-side
constructor.
1. Fixed a design flaw in PocketManager. We originally assumed that all
requests to PocketManager.getDimensionData() had to be legitimate
requests for dimensions that existed. That was true in most cases, but
for things like processing user commands, it was dangerously optimistic.
It was possible that a flaw in DD's usage of that function could be
exploited by a player to trick the mod into pre-registering dimension
data for a non-existent dimension. That would declare the dimension as a
root. DD would crash later if Forge ever allocated that ID for a pocket
dimension. The new implementation is almost the same as the old one, but
allows us to differentiate between cases when we can eagerly create
dimension data, and cases in which the absence of a dimension should
cause a crash to alert us of a design flaw.
2. Remove the pocket regeneration code from PocketBuilder. We simply
don't support pocket regeneration and it's unlikely it'll ever be
implemented because it's a difficult issue. Wiping out pockets
completely is easier. We can always recover the code from this commit if
it's needed later.
3. Minor changes: removed some debug prints from PocketManager and
changed some static accesses in PocketBuilder.
Fixed an issue. DD would crash when MC created a completely new world
because onChunkLoad() would be called before onWorldLoad(). That's not
the usual order. PocketManager would be unloaded at that point and would
return a null dimension.
Autocorrected indentation in PocketManager. I'll be working on changing
PocketManager to prevent the risk of creating data for a non-existent
dimension through a bad call to getDimensionData().
Made some improvements to TileEntityRift. The main reason for these
changes was to remove the field nearestRift - we should not hold on to
references to links. Now we simply track the location of the nearest
rift. I also confirmed that closeRift() and updateNearestRift() must be
allowed to run on both the client and the server. If the client doesn't
run those functions, then adjacent rifts don't connect as expected and
the rift removal animation doesn't work.
Removed a pointless check in yCoordHelper and corrected some comments in
RiftRegenerator. It turns out that ChunkProviderServer.chunkExists()
returns whether a chunk is loaded, not whether it has already been
created.
Removed unnecessary code for the Transdimensional Trapdoor. Most of it
was code dedicated to updating TileEntityTransTrapdoor.hasRift. That
flag was never used for anything.
1. Added code so that Transdimensional Trapdoors detect that they have
been broken and schedule rift regeneration at their location. This had
previously been neglected. Trapdoors deserve a little more attention.
2. Tweaked the breakBlock() code for BlockRift and BaseDimDoor so that
rift regeneration is only scheduled if the underlying block was removed.
We don't want that to happen if the only change was for metadata.
1. Removed code from BaseDimDoor that was already implemented almost
identically in BlockDoor. Clarified some uses of setBlock() by changing
them to setBlockToAir() instead.
2. Removed TileEntityDimDoor.invalidate() and moved the regeneration
scheduling code to BaseDimDoor.breakBlock(). I would prefer to move away
from overriding the invalidate() method. This also simplifies the code
since we don't need to perform some of the checks we had in
breakBlock().
1. Made it so that rifts regenerate when rift blocks are replaced by
other blocks.
2. Changed the rift regeneration scheduling functions to streamline
their use in other classes. Common code that was needed to validate
links before calling those functions has been moved into them so that
the checks are always performed internally.
1. Implemented scheduled rift regeneration in RiftRegenerator. The
previous randomized selection algorithm has been removed completely. All
regeneration is scheduled now. We perform numerous checks to make sure
that regenerating a rift is safe.
2. Removed FastRiftRegenerator as RiftRegenerator performs roughly the
same task but with more flexibility. Updated TileEntityDimDoor to use
RiftRegenerator instead for creating rifts when doors are broken.
3. Modified EventHookContainer to receive the chunk loaded event. We
iterate over the list of links in a loaded chunk and schedule them for
regeneration.
4. Reorganized the code in BlockRift. Divided the list of immune blocks
into two lists - one for DD blocks and one for regular MC blocks.
RiftRegenerator has to be able to distinguish between the two types.
5. Factored out some duplicate code from ItemRiftSignature and
ItemStabilizedRiftSignature. Most of the block immunity checks were used
to check if it would be safe to spawn a rift when using one of those
items. BlockRift.tryPlacingRift() covers all that logic in a single
function and makes the item code a little simpler.
Minor changes to DimLink to simplify DimLink.getDestinationOrientation()
and to clarify the output of DimLink.toString() when no destination is
available.
Rewrote the NewDimData.deleteLink() version that would accept x, y, and
z as parameters. There was some redundant code for getting the
parameters from a Point4D instance just to create another one to find
the target link. Now we pass the source point in directly.
Made some minor changes to NewDimData. Fixed some comments for the rift
search functions that incorrectly indicated the search would not detect
rifts adjacent to the center of the search range. That behavior changed
some time ago. Also added an unused field called "chunkMapping" for
associating chunks with lists of links. It'll be used for implementing
queuing of rift regeneration when chunks load.
1. Reorganized our code to initialize tick receivers each time the
server starts rather than once when the mod is initialized. This is
needed because reusing a single instance of each class across different
single-player sessions could cause scheduled events for one world to
leak into another world. This approach ensures that we discard all
pending events.
2. Separated the implementation of Limbo decay from a tick receiver that
periodically triggers fast decay. All of the decay code has been kept in
LimboDecay, while the ticking is handled by LimboDecayScheduler. This
change separates some functionality that should be independent, but
also, it's needed so that BlockLimbo can have access to LimboDecay's
methods without holding on to a tick receiver instance.
3. Minor change: renamed ChunkLoaderHelper.loadChunkForcedWorlds() to
loadForcedChunkWorlds().
