mirror of
https://github.com/sbrl/Minetest-WorldEditAdditions.git
synced 2024-11-16 13:03:00 +00:00
Starbeamrainbowlabs
fd5804dd9c
Specifically, I'm unsure about whether I'm happy with the effects of the algorithm. Also, we convolve with a 3x3 gaussian kernel after erosion is complete - and we have verified that the erosion is having an positive effect at "roughening up" a terrain surface. It seems like the initial blog post was correct: the algorithm does tend to make steep surfaces steeper. It also appears that it's more effective on larger areas, and 'gentler' curves. THis might be because the surface normals are more conducive to making the snowballs roll. Finally, we need to decide whether we want to keep the precomputed normals as we have now, or whether we want to dynamically compute them at the some of request.
264 lines
8.3 KiB
Lua
264 lines
8.3 KiB
Lua
-- Licence: GPLv2 (MPL-2.0 is compatible, so we can use this here)
|
|
-- Source: https://stackoverflow.com/a/43582076/1460422
|
|
|
|
-- gsplit: iterate over substrings in a string separated by a pattern
|
|
--
|
|
-- Parameters:
|
|
-- text (string) - the string to iterate over
|
|
-- pattern (string) - the separator pattern
|
|
-- plain (boolean) - if true (or truthy), pattern is interpreted as a plain
|
|
-- string, not a Lua pattern
|
|
--
|
|
-- Returns: iterator
|
|
--
|
|
-- Usage:
|
|
-- for substr in gsplit(text, pattern, plain) do
|
|
-- doSomething(substr)
|
|
-- end
|
|
function worldeditadditions.gsplit(text, pattern, plain)
|
|
local splitStart, length = 1, #text
|
|
return function ()
|
|
if splitStart then
|
|
local sepStart, sepEnd = string.find(text, pattern, splitStart, plain)
|
|
local ret
|
|
if not sepStart then
|
|
ret = string.sub(text, splitStart)
|
|
splitStart = nil
|
|
elseif sepEnd < sepStart then
|
|
-- Empty separator!
|
|
ret = string.sub(text, splitStart, sepStart)
|
|
if sepStart < length then
|
|
splitStart = sepStart + 1
|
|
else
|
|
splitStart = nil
|
|
end
|
|
else
|
|
ret = sepStart > splitStart and string.sub(text, splitStart, sepStart - 1) or ''
|
|
splitStart = sepEnd + 1
|
|
end
|
|
return ret
|
|
end
|
|
end
|
|
end
|
|
|
|
|
|
-- split: split a string into substrings separated by a pattern.
|
|
--
|
|
-- Parameters:
|
|
-- text (string) - the string to iterate over
|
|
-- pattern (string) - the separator pattern
|
|
-- plain (boolean) - if true (or truthy), pattern is interpreted as a plain
|
|
-- string, not a Lua pattern
|
|
--
|
|
-- Returns: table (a sequence table containing the substrings)
|
|
function worldeditadditions.split(text, pattern, plain)
|
|
local ret = {}
|
|
for match in worldeditadditions.gsplit(text, pattern, plain) do
|
|
table.insert(ret, match)
|
|
end
|
|
return ret
|
|
end
|
|
|
|
|
|
--- Pads str to length len with char from right
|
|
-- @source https://snipplr.com/view/13092/strlpad--pad-string-to-the-left
|
|
function worldeditadditions.str_padend(str, len, char)
|
|
if char == nil then char = ' ' end
|
|
return str .. string.rep(char, len - #str)
|
|
end
|
|
--- Pads str to length len with char from left
|
|
-- Adapted from the above
|
|
function worldeditadditions.str_padstart(str, len, char)
|
|
if char == nil then char = ' ' end
|
|
return string.rep(char, len - #str) .. str
|
|
end
|
|
|
|
--- Equivalent to string.startsWith in JS
|
|
-- @param str string The string to operate on
|
|
-- @param start number The start string to look for
|
|
-- @returns bool Whether start is present at the beginning of str
|
|
function worldeditadditions.string_starts(str,start)
|
|
return string.sub(str,1,string.len(start))==start
|
|
end
|
|
|
|
--- Prints a 2d array of numbers formatted like a JS TypedArray (e.g. like a manip node list or a convolutional kernel)
|
|
-- In other words, the numbers should be formatted as a single flat array.
|
|
-- @param tbl number[] The ZERO-indexed list of numbers
|
|
-- @param width number The width of 2D array.
|
|
function worldeditadditions.print_2d(tbl, width)
|
|
print("==== count: "..#tbl..", width:"..width.." ====")
|
|
local display_width = 1
|
|
for _i,value in pairs(tbl) do
|
|
display_width = math.max(display_width, #tostring(value))
|
|
end
|
|
display_width = display_width + 2
|
|
local next = {}
|
|
for i=0, #tbl do
|
|
table.insert(next, worldeditadditions.str_padstart(tostring(tbl[i]), display_width))
|
|
if #next == width then
|
|
print(table.concat(next, ""))
|
|
next = {}
|
|
end
|
|
end
|
|
end
|
|
|
|
|
|
|
|
--- Turns an associative node_id → count table into a human-readable list.
|
|
-- Works well with worldeditadditions.make_ascii_table().
|
|
function worldeditadditions.node_distribution_to_list(distribution, nodes_total)
|
|
local distribution_data = {}
|
|
for node_id, count in pairs(distribution) do
|
|
table.insert(distribution_data, {
|
|
count,
|
|
tostring(worldeditadditions.round((count / nodes_total) * 100, 2)).."%",
|
|
minetest.get_name_from_content_id(node_id)
|
|
})
|
|
end
|
|
return distribution_data
|
|
end
|
|
|
|
--- Makes a human-readable table of data.
|
|
-- Data should be a 2D array - i.e. a table of tables. The nested tables should
|
|
-- contain a list of items for a single row.
|
|
-- If total is specified, then a grand total is printed at the bottom - this is
|
|
-- useful when you want to print a node list - works well with
|
|
-- worldeditadditions.node_distribution_to_list().
|
|
function worldeditadditions.make_ascii_table(data, total)
|
|
local extra_padding = 2
|
|
local result = {}
|
|
local max_lengths = {}
|
|
for y = 1, #data, 1 do
|
|
for x = 1, #data[y], 1 do
|
|
if not max_lengths[x] then
|
|
max_lengths[x] = 0
|
|
end
|
|
max_lengths[x] = math.max(max_lengths[x], #tostring(data[y][x]) + extra_padding)
|
|
end
|
|
end
|
|
|
|
for _key, row in ipairs(data) do
|
|
local row_result = {}
|
|
for i = 1, #row, 1 do
|
|
row_result[#row_result + 1] = worldeditadditions.str_padend(tostring(row[i]), max_lengths[i], " ")
|
|
end
|
|
result[#result+1] = table.concat(row_result, "")
|
|
end
|
|
|
|
if total then
|
|
result[#result+1] = string.rep("=", 6 + #tostring(total) + 6).."\n"..
|
|
"Total "..total.." nodes\n"
|
|
end
|
|
|
|
-- TODO: Add multi-column support here
|
|
return table.concat(result, "\n")
|
|
end
|
|
|
|
--- Parses a list of strings as a list of weighted nodes - e.g. like in the //mix command.
|
|
-- @param parts string[] The list of strings to parse (try worldeditadditions.split)
|
|
-- @param as_list bool If true, then table.insert() successive { node = string, weight = number } subtables when parsing instead of populating as an associative array.
|
|
-- @returns table A table in the form node_name => weight.
|
|
function worldeditadditions.parse_weighted_nodes(parts, as_list)
|
|
if as_list == nil then as_list = false end
|
|
local MODE_EITHER = 1
|
|
local MODE_NODE = 2
|
|
|
|
local result = {}
|
|
local mode = MODE_NODE
|
|
local last_node_name = nil
|
|
for i, part in ipairs(parts) do
|
|
print("i: "..i..", part: "..part)
|
|
if mode == MODE_NODE then
|
|
print("mode: node");
|
|
local next = worldedit.normalize_nodename(part)
|
|
if not next then
|
|
return false, "Error: Invalid node name '"..part.."'"
|
|
end
|
|
last_node_name = next
|
|
mode = MODE_EITHER
|
|
elseif mode == MODE_EITHER then
|
|
print("mode: either");
|
|
local chance = tonumber(part)
|
|
if not chance then
|
|
print("not a chance, trying a node name")
|
|
local node_name = worldedit.normalize_nodename(part)
|
|
if not node_name then
|
|
return false, "Error: Invalid number '"..chance.."'"
|
|
end
|
|
if last_node_name then
|
|
if as_list then table.insert(result, { node = last_node_name, weight = 1 })
|
|
else result[last_node_name] = 1 end
|
|
end
|
|
last_node_name = node_name
|
|
mode = MODE_EITHER
|
|
else
|
|
print("it's a chance: ", chance, "for", last_node_name)
|
|
chance = math.floor(chance)
|
|
if as_list then table.insert(result, { node = last_node_name, weight = chance })
|
|
else result[last_node_name] = chance end
|
|
last_node_name = nil
|
|
mode = MODE_NODE
|
|
end
|
|
end
|
|
end
|
|
if last_node_name then
|
|
print("caught trailing node name: ", last_node_name)
|
|
if as_list then table.insert(result, { node = last_node_name, weight = 1 })
|
|
else result[last_node_name] = 1 end
|
|
end
|
|
|
|
return true, result
|
|
end
|
|
|
|
function worldeditadditions.parse_map(params_text)
|
|
local result = {}
|
|
local parts = worldeditadditions.split(params_text, "%s+", false)
|
|
|
|
local last_key = nil
|
|
for i, part in ipairs(parts) do
|
|
if i % 2 == 0 then -- Lua starts at 1 :-/
|
|
-- Try converting to a number to see if it works
|
|
local part_converted = tonumber(part)
|
|
if as_number == nil then part_converted = part end
|
|
result[last_key] = part
|
|
else
|
|
last_key = part
|
|
end
|
|
end
|
|
return true, result
|
|
end
|
|
|
|
function worldeditadditions.map_stringify(map)
|
|
local result = {}
|
|
for key, value in pairs(map) do
|
|
table.insert(result, key.."\t"..value)
|
|
end
|
|
return table.concat(result, "\n")
|
|
end
|
|
|
|
--- Converts a float milliseconds into a human-readable string.
|
|
-- Ported from PHP human_time from Pepperminty Wiki: https://github.com/sbrl/Pepperminty-Wiki/blob/fa81f0d/core/05-functions.php#L82-L104
|
|
-- @param ms float The number of milliseconds to convert.
|
|
-- @return string A human-readable string representing the input ms.
|
|
function worldeditadditions.human_time(ms)
|
|
local tokens = {
|
|
{ 31536000 * 1000, 'year' },
|
|
{ 2592000 * 1000, 'month' },
|
|
{ 604800 * 1000, 'week' },
|
|
{ 86400 * 1000, 'day' },
|
|
{ 3600 * 1000, 'hr' },
|
|
{ 60 * 1000, 'min' },
|
|
{ 1 * 1000, 's' },
|
|
{ 1, 'ms'}
|
|
}
|
|
|
|
for _,pair in pairs(tokens) do
|
|
if ms > pair[1] or pair[2] == "ms" then
|
|
local unit = pair[2]
|
|
if ms > 60 * 1000 and math.floor(ms / pair[1]) > 1 then
|
|
unit = unit.."s"
|
|
end
|
|
return string.format("%.2f", ms / pair[1])..unit
|
|
end
|
|
end
|
|
end
|