Scrub mod manifest strings to valid UTF-8

A manifest whose name, version, description, or category carries invalid
UTF-8 (a BOM, Latin-1 bytes) crashed the launcher's MODS panel, since
love.graphics.printf raises on invalid UTF-8. Manifest.validate now drops
invalid bytes and a leading BOM from those strings, in place so the
badge's raw.category read agrees.
This commit is contained in:
bryanthaboi
2026-08-01 14:22:26 -04:00
parent d26d63ed38
commit 6bb2e078c0
2 changed files with 81 additions and 0 deletions
+50
View File
@@ -86,8 +86,58 @@ local function mergeConflictLists(conflicts, incompatible)
return out
end
-- Drop bytes that are not valid UTF-8 (malformed sequences, overlongs,
-- surrogates, > U+10FFFF) and a leading BOM. LÖVE's text renderer raises
-- "Invalid UTF-8" from love.graphics.print/printf, so any manifest string a
-- panel may draw must be scrubbed here -- the one place every mod manifest
-- passes through -- or a single mangled description crashes the whole MODS
-- panel instead of misrendering one card.
local function scrubUtf8(s)
if type(s) ~= "string" then return s end
s = s:gsub("^\239\187\191", "")
local out, i, n = {}, 1, #s
while i <= n do
local b = s:byte(i)
local len
if b < 0x80 then len = 1
elseif b >= 0xC2 and b <= 0xDF then len = 2
elseif b >= 0xE0 and b <= 0xEF then len = 3
elseif b >= 0xF0 and b <= 0xF4 then len = 4
end
local ok = len ~= nil and i + len - 1 <= n
if ok and len > 1 then
for j = i + 1, i + len - 1 do
local c = s:byte(j)
if c < 0x80 or c > 0xBF then ok = false; break end
end
if ok then
-- boundary lead bytes narrow their second byte: no overlongs
-- (E0/F0), no surrogates (ED), nothing past U+10FFFF (F4)
local b2 = s:byte(i + 1)
if (b == 0xE0 and b2 < 0xA0) or (b == 0xED and b2 > 0x9F)
or (b == 0xF0 and b2 < 0x90) or (b == 0xF4 and b2 > 0x8F) then
ok = false
end
end
end
if ok then
out[#out + 1] = s:sub(i, i + len - 1)
i = i + len
else
i = i + 1
end
end
return table.concat(out)
end
function Manifest.validate(raw, path)
assert(type(raw) == "table", "manifest must be an object")
-- scrubbed in place so every later reader agrees, including the launcher's
-- badge derivation, which reads raw.category rather than the validated copy
raw.name = scrubUtf8(raw.name)
raw.version = scrubUtf8(raw.version)
raw.description = scrubUtf8(raw.description)
raw.category = scrubUtf8(raw.category)
assert(type(raw.id) == "string" and raw.id:match("^[%w_%-]+$"),
"manifest id must contain only letters, numbers, _ or -")
assert(type(raw.name) == "string" and raw.name ~= "", "manifest name is required")