This reverts commit 13d2ae20
Why was this changed. I doubt it did much in terms of performance. It's better to stay as close as we can to the worldedit repo.
* Sync up implementations of the two commands.
* Fix generating trees in spots with replaceable blocks.
* Make message when you mistype tree-type arg more correct.
Fixes WORLDEDIT-3869.
Previously, the current request would just get a new EditSession when
one was created. Now, a Request is reset before and after:
- a command is used and
- an interact is fired with the platform
This means each action taken will get a single, non-reusable Request.
Note that this only applies to actions taken through the platform.
API users will not be using requests anyway, since things like Masks,
etc. will be constructed directly instead of being passed through the
platform's parsers and so on. (e.g. if a plugin loads a schematic into
the world with a mask, they should create the EditSession and mask it
directly, and not use that Mask again for another EditSession in another
World).
Also, get rid of a bunch of (some now-)unnecessary EditSession creation
during command dispatching.
Note that this also fixed the dynamic selection mask, which apparently
has been broken for some unknown amount of time.
For example, if you set a mask that takes an extent (many of them),
and then move to another world, the mask will test blocks in the old
world and return bad results.
* Not all EntityTypes in Bukkit have the correct enum name.
* Don't read entire schematic files to list. Go off file extension only.
(Reading in files is more accurate, but slow.)
* Enforce extensions. (Due to the above, while you can technically make a
schematic called 'test.txt', it's better that we save it as
'test.txt.schem'.)
* Fix a few minor warnings.
Add better control over expression timeouts.
* //timeout command can be used to change player's current timeout.
* Config now also has a max timeout, can be bypassed with permission
* Timeout of < 0 will let expressions run indefinitely.
* Said expressions won't run on a separate thread, slightly reducing the
overhead from context switching. For large //gen commands, for example,
this can actually increase speed.
Instead of trying to maintain a list of "natural terrain blocks", just
let the user specify a mask of blocks to use for the height map filter.
https://gfycat.com/severaljauntycondor
The initial material is held in the factory, but if a brush was previously bound, that pattern is used.
Now, using `//br set` will clear the previous material, allowing the OperationFactory's material to work.
This can be changed later with `/mat`, which will once again set the fill on the tool, overriding the factory's context.
* `//set ##*tag` sets all states in the tag (not just default state per type)
* `//set ^type` is a pattern changing block type but copying all valid existing states
* `//set ^[prop=val,...]` sets the property `prop` to `val` wherever the existing block has that property
* `//set ^type[prop=val,...]` does both of the above
Those work anywhere a pattern is taken, of course.
* The mask syntax `^[prop=val]` matches blocks with the property `prop` set to `val`, or blocks that don't have the property at all.
* The mask syntax `^=[prop=val]` only matches blocks that have the property.
Those work anywhere a mask is taken, of course. (`//mask`, `//gmask`, `//replace`, etc)
The `//drain` command now takes `-w` flag that removes the waterlogged state from blocks (in addition to removing water, as before).
Unfortunately this is a bit slower than before since we can't cache block
id & data values. However, applying patterns generally isn't too expensive,
and hollow regions were entirely broken before.
