mirror of
https://github.com/appgurueu/modlib.git
synced 2024-11-26 01:03:46 +01:00
Text utility cleanup
This commit is contained in:
parent
1e4f795691
commit
8a6c22d107
138
text.lua
138
text.lua
@ -1,123 +1,116 @@
|
||||
-- String helpers - split & trim at end & begin
|
||||
function upper_first(str)
|
||||
return str:sub(1,1):upper()..str:sub(2)
|
||||
end
|
||||
function lower_first(str)
|
||||
return str:sub(1,1):lower()..str:sub(2)
|
||||
end
|
||||
function starts_with(str, start) return str:sub(1, start:len()) == start end
|
||||
function ends_with(str, suffix)
|
||||
return str:sub(str:len() - suffix:len() + 1) == suffix
|
||||
end
|
||||
function trim(str, to_remove)
|
||||
function upper_first(text) return text:sub(1, 1):upper() .. text:sub(2) end
|
||||
|
||||
function lower_first(text) return text:sub(1, 1):lower() .. text:sub(2) end
|
||||
|
||||
function starts_with(text, start) return text:sub(1, start:len()) == start end
|
||||
|
||||
function ends_with(text, suffix) return text:sub(text:len() - suffix:len() + 1) == suffix end
|
||||
|
||||
function trim(text, to_remove)
|
||||
local j = 1
|
||||
for i = 1, string.len(str) do
|
||||
if str:sub(i, i) ~= to_remove then
|
||||
for i = 1, string.len(text) do
|
||||
if text:sub(i, i) ~= to_remove then
|
||||
j = i
|
||||
break
|
||||
end
|
||||
end
|
||||
|
||||
local k = 1
|
||||
for i = string.len(str), j, -1 do
|
||||
if str:sub(i, i) ~= to_remove then
|
||||
for i = string.len(text), j, -1 do
|
||||
if text:sub(i, i) ~= to_remove then
|
||||
k = i
|
||||
break
|
||||
end
|
||||
end
|
||||
|
||||
return str:sub(j, k)
|
||||
return text:sub(j, k)
|
||||
end
|
||||
|
||||
function trim_begin(str, to_remove)
|
||||
function trim_begin(text, to_remove)
|
||||
local j = 1
|
||||
for i = 1, string.len(str) do
|
||||
if str:sub(i, i) ~= to_remove then
|
||||
for i = 1, string.len(text) do
|
||||
if text:sub(i, i) ~= to_remove then
|
||||
j = i
|
||||
break
|
||||
end
|
||||
end
|
||||
|
||||
return str:sub(j)
|
||||
return text:sub(j)
|
||||
end
|
||||
|
||||
trim_left = trim_begin
|
||||
|
||||
function trim_end(str, to_remove)
|
||||
function trim_end(text, to_remove)
|
||||
local k = 1
|
||||
for i = string.len(str), 1, -1 do
|
||||
if str:sub(i, i) ~= to_remove then
|
||||
for i = string.len(text), 1, -1 do
|
||||
if text:sub(i, i) ~= to_remove then
|
||||
k = i
|
||||
break
|
||||
end
|
||||
end
|
||||
return str:sub(1, k)
|
||||
return text:sub(1, k)
|
||||
end
|
||||
|
||||
trim_right = trim_end
|
||||
|
||||
function split(str, delim, limit, regex)
|
||||
if not limit then return split_without_limit(str, delim, regex) end
|
||||
local no_regex = not regex
|
||||
function split(text, delimiter, limit, is_regex)
|
||||
limit = limit or math.huge
|
||||
local no_regex = not is_regex
|
||||
local parts = {}
|
||||
local occurences = 1
|
||||
local last_index = 1
|
||||
local index = string.find(str, delim, 1, no_regex)
|
||||
local index = string.find(text, delimiter, 1, no_regex)
|
||||
while index and occurences < limit do
|
||||
table.insert(parts, string.sub(str, last_index, index - 1))
|
||||
last_index = index + string.len(delim)
|
||||
index = string.find(str, delim, index + string.len(delim), no_regex)
|
||||
table.insert(parts, string.sub(text, last_index, index - 1))
|
||||
last_index = index + string.len(delimiter)
|
||||
index = string.find(text, delimiter, index + string.len(delimiter), no_regex)
|
||||
occurences = occurences + 1
|
||||
end
|
||||
table.insert(parts, string.sub(str, last_index))
|
||||
table.insert(parts, string.sub(text, last_index))
|
||||
return parts
|
||||
end
|
||||
|
||||
function split_without_limit(str, delim, regex)
|
||||
local no_regex = not regex
|
||||
local parts = {}
|
||||
local last_index = 1
|
||||
local index = string.find(str, delim, 1, no_regex)
|
||||
while index do
|
||||
table.insert(parts, string.sub(str, last_index, index - 1))
|
||||
last_index = index + string.len(delim)
|
||||
index = string.find(str, delim, index + string.len(delim), no_regex)
|
||||
end
|
||||
table.insert(parts, string.sub(str, last_index))
|
||||
return parts
|
||||
end
|
||||
function split_without_limit(text, delimiter, is_regex) return split(text, delimiter, nil, is_regex) end
|
||||
|
||||
split_unlimited = split_without_limit
|
||||
|
||||
function split_lines(str, limit)
|
||||
modlib.text.split(str, "\r?\n", limit, true)
|
||||
end
|
||||
function split_lines(text, limit) return modlib.text.split(text, "\r?\n", limit, true) end
|
||||
|
||||
hashtag = string.byte("#")
|
||||
zero = string.byte("0")
|
||||
nine = string.byte("9")
|
||||
letter_a = string.byte("A")
|
||||
letter_f = string.byte("F")
|
||||
function lines(text) return text:gmatch"[^\r\n]*" end
|
||||
|
||||
local hashtag = string.byte"#"
|
||||
local zero = string.byte"0"
|
||||
local nine = string.byte"9"
|
||||
local letter_a = string.byte"A"
|
||||
local letter_f = string.byte"F"
|
||||
|
||||
function is_hexadecimal(byte)
|
||||
return (byte >= zero and byte <= nine) or
|
||||
(byte >= letter_a and byte <= letter_f)
|
||||
return byte >= zero and byte <= nine or byte >= letter_a and byte <= letter_f
|
||||
end
|
||||
|
||||
magic_chars = {
|
||||
"%", "(", ")", ".", "+", "-", "*", "?", "[", "^", "$" --[[,":"]]
|
||||
"%",
|
||||
"(",
|
||||
")",
|
||||
".",
|
||||
"+",
|
||||
"-",
|
||||
"*",
|
||||
"?",
|
||||
"[",
|
||||
"^",
|
||||
"$"
|
||||
}
|
||||
local magic_charset = {}
|
||||
for _, magic_char in pairs(magic_chars) do table.insert(magic_charset, "%" .. magic_char) end
|
||||
magic_charset = "[" .. table.concat(magic_charset) .. "]"
|
||||
|
||||
function escape_magic_chars(text)
|
||||
for _, magic_char in ipairs(magic_chars) do
|
||||
text = string.gsub(text, "%" .. magic_char, "%%" .. magic_char)
|
||||
end
|
||||
return text
|
||||
end
|
||||
function escape_magic_chars(text) return text:gsub("(" .. magic_charset .. ")", "%%%1") end
|
||||
|
||||
function utf8(number)
|
||||
if number < 0x007F then return string.char(number) end
|
||||
if number < 0x00A0 or number > 0x10FFFF then -- Out of range
|
||||
if number < 0x007F then
|
||||
-- Single byte
|
||||
return string.char(number)
|
||||
end
|
||||
if number < 0x00A0 or number > 0x10FFFF then
|
||||
-- Out of range
|
||||
return
|
||||
end
|
||||
local result = ""
|
||||
@ -129,17 +122,17 @@ function utf8(number)
|
||||
i = i + 1
|
||||
if number <= math.pow(2, 8 - i - 2) then break end
|
||||
end
|
||||
return string.char(256 - math.pow(2, 8 - i - 1) + number) .. result -- 256 = math.pow(2, 8)
|
||||
return string.char(256 - math.pow(2, 8 - i - 1) + number) .. result
|
||||
end
|
||||
|
||||
--+ deprecated
|
||||
function handle_ifdefs(code, vars)
|
||||
local finalcode = {}
|
||||
local endif
|
||||
local after_endif = -1
|
||||
local ifdef_pos, after_ifdef = string.find(code, "--IFDEF", 1, true)
|
||||
while ifdef_pos do
|
||||
table.insert(finalcode,
|
||||
string.sub(code, after_endif + 2, ifdef_pos - 1))
|
||||
table.insert(finalcode, string.sub(code, after_endif + 2, ifdef_pos - 1))
|
||||
local linebreak = string.find(code, "\n", after_ifdef + 1, true)
|
||||
local varname = string.sub(code, after_ifdef + 2, linebreak - 1)
|
||||
endif, after_endif = string.find(code, "--ENDIF", linebreak + 1, true)
|
||||
@ -147,8 +140,7 @@ function handle_ifdefs(code, vars)
|
||||
if vars[varname] then
|
||||
table.insert(finalcode, string.sub(code, linebreak + 1, endif - 1))
|
||||
end
|
||||
ifdef_pos, after_ifdef = string.find(code, "--IFDEF",
|
||||
after_endif + 1, true)
|
||||
ifdef_pos, after_ifdef = string.find(code, "--IFDEF", after_endif + 1, true)
|
||||
end
|
||||
table.insert(finalcode, string.sub(code, after_endif + 2))
|
||||
return table.concat(finalcode, "")
|
||||
|
Loading…
Reference in New Issue
Block a user