Fix for Vector/BlockVector
Use optimized LocalBlockVectorSet for various operations
- Ideally I'd like the operations to use O(1) memory, but for now it'll
use substantially reduced O(n)
Some message tweaks
Some minor entity placement optimizations
Refactor set optimization to be in the EditSession class
Fix and minor optimizations for for countBlocks()
Minor optimizations for affine transforms (//rotate etc.)
Optimizations for fuzzy block mask
Added command tab completion for patterns
Tick limiter:
- Ignore redstone physics
- Per chunk limiting (rather than global)
- Notify console of any limiting
Changes default settings for low memory usage rather than speed:
- Doesn't effect any existing configs, just new installs
- Uses disk and database for undo/redo/clipboard
- Uses compression level 8 instead of 1
Fixes 1.8 queue failing on no isDirty field
Fixes rare chunk skipping
Fixes queue staging issue
Tweak undo/redo failure messages to provide more info
Added sand/gravel sphere message so that people don't ask me "why it no
work!?"
Add #clipboard transform
Add #fullcopy pattern
- similar to transform, it pastes the full clipboard at any changed
block
- e.g. //replace <marker> #fullcopy
Async block get optimizations for bukkit by running tasks multiple times
during a single tick
Tweak the OOM message to be more informative
Tweak the max memory config comment to be more informative
Restructured transforms to use resettable extent + reduce code
duplication
Clipboards can now be streamed to a schematic file without significant
memory overhead
- This means you can now load/paste/copy/save arbitrarily large sizes
with fixed memory usage
Optimizations to the various clipboard implementations
Add optimized extent block translation (no additional object creation
for set block)
Optimized forward extent block copy
- use optimized extent transform/translate
- pre calculate required functions/extents outside iterations
- short circuit certain functions depending on input parameters
Use edit session for patterns rather than world (faster/safer)
Schematics now load in a fraction of a time. (took me 5.13 seconds to
load a 102,572,228 block schematic)
Also made some minor optimizations here and there.
The CuboidRegion class will now queue blocks in layers for a chunk
before moving onto the next chunk. This results in higher cache hits for
history enabled queues. It also allows the block placer to start earlier
during preprocessing with edits affecting > 64 (configurable) chunks.
Note: with history on disk enabled, this means near unlimited sized
edits (for certain commands) might be feasible.
fix a few async issues (nbt setting/cut/plotsquared hook)
fixes#1
allow some debug options to be configurable (chunk wait / buffer size)
add region restriction option
Using java's inbuilt compression algorithms (gzip or deflator both) was
a bit too slow for me.
Fawe is now using the LZ4 compression algorithm, which is super fast. So
fast that I decided to add a default 1 pass of compression to history.
Enabling further compression will do an additional high LZ4 compression.
Streamlined the queue so now the queue doesn't need to be fetched on
each block
Removed block reording from editsession as it alone takes half a
second/million blocks
Replaced block fetch from history extent with optimized algorithm.
