For some specific out of bounds values, the volume calculation in
minetest.find_nodes_in_area() is off by about four million nodes.
Unfortunately that behaviour made it trivial to crash Mineclonia,
as Minetest immediately crashes upon encountering large numbers.
This commit introduces a wrapper around minetest.find_nodes_in_area()
which should avoid a crash. Additionally, three self tests are executed
when a server starts; they crash Mineclonia in case the workaround fails.
Without this fix, the banner pattern preview generation does not mask
the banner pattern, so the alpha channel of the banner pattern is not
taken into account. This lead to preview banners with color gradients
showing up as a solid color banner and opaque pixel artifacts for the
bottom triangle pattern.
Though usually one would export directly to "${3}", Inkscape 1.0 had
its command line options changed by people who apparently think that
backwards compatibility is some kind of swear word: Whereas earlier
Inkscape versions would export to a file called foo.png.tmp, newer
behaviour is to ignore the user's wishes & write to foo.png.png –
unless one asks it to write to a filename with a .png extension,
Inkscape 1.0 changes the filename extension to .png each time.
As we do not know the extension of "${3}", we have to use the
extension, then rename the resulting file to the proper name;
only that way the export works with Inkscape 1.0 & earlier …
When a player joins and immediately leaves the game before a function is
called by minetest.after() in mods/PLAYER/wieldview/init.lua, it gets an
invalidated player object. This results in the player methods returning
nil (since Minetest 5.2); perhaps surprisingly, the player is not nil.
Not checking that the result of player:get_pos() is not nil could lead
to a server crash if a client crashed when joining. It has been reported
that a syntax error in a client side mod was enough to trigger that.
The “getwrittenbook” command gives a player that has the “debug” privilege a book
with a configurable amount of characters. This was added as a debug aid, to help
reproducing situations in which items with lots of metadata trigger issues like
heavy lag or server crashes.
Some items, like shulkers or books, can have so much metadata that the
corresponding item entity can not be serialized by the Minetest engine.
Without this patch, dropping such an item and then moving away crashes
Minetest, as it can not serialize the entity with serializeString16()
when unloading a map block.
The patch resets the overlong metadata of non-serializable item entities.
This avoids a crash and makes it possible to retrieve a “sanitized” item
without metadata when the mapblock containing the item entity is reloaded.
Originally sfan5 guessed the maximum possible item entity serialization size
that would not lead to a crash as 65530 bytes, but anon5 calculated it to be
actually 65487 bytes. This has been experimentally verified by erlehmann.
In Minecraft Java Edition, when the map generator generates a spawner,
it can generates a pig spawner instead of the spawner it should create.
That behaviour is very rare, but has never been removed from Minecraft.
This patch changes 1 in 1000 spawners in a mineshaft to be a pig spawner
instead of a cave spider spawner.
Several mods set or unset the visibility of a HUD bar way too often (e.g.
in a globalstep handler), causing the server to send a lot of superfluous
HUDCHANGE packets to each client. Returning from hb.hide_hudbar() early
if HUD bar visibility would not change prevents sending these packets.
Mineclonia has inherited mods from MineClone 2 that send a lot of network
packets. This behaviour wastes bandwith and is most likely a major reason
for the unusually high amount of lag that MineClone2 and Mineclonia have.
Many network packets that are sent by Mineclonia are entirely useless.
Analyzing minetest log files to figure out what kind of packets are
sent and how often is a first step in getting rid of useless traffic.
