From 6bb2e078c05b774f614f0c3dcd1fef8c256919fd Mon Sep 17 00:00:00 2001
From: bryanthaboi <bryanthaboi@theboisclub.com>
Date: Sat, 1 Aug 2026 14:22:26 -0400
Subject: [PATCH] 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.
---
 src/mods/Manifest.lua                | 50 ++++++++++++++++++++++++++++
 tests/engine/launcher_mods_tests.lua | 31 +++++++++++++++++
 2 files changed, 81 insertions(+)

diff --git a/src/mods/Manifest.lua b/src/mods/Manifest.lua
index d1f1545b..12ff83a7 100644
--- a/src/mods/Manifest.lua
+++ b/src/mods/Manifest.lua
@@ -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")
diff --git a/tests/engine/launcher_mods_tests.lua b/tests/engine/launcher_mods_tests.lua
index d2a2a5f0..02118482 100644
--- a/tests/engine/launcher_mods_tests.lua
+++ b/tests/engine/launcher_mods_tests.lua
@@ -297,4 +297,35 @@ do
   eq(rows[1].name, "bare", "a nameless row falls back to its id")
 end
 
+-- ------- manifest strings are scrubbed to valid UTF-8 (MODS panel crash:
+-- LÖVE's printf raises "Invalid UTF-8" on a mangled name/description, so
+-- validate must drop bad bytes before any panel draws them)
+
+do
+  local m = mf({ id = "utf", entry = "m.lua",
+    -- BOM-prefixed name (a real manifest shipped this way), a Latin-1 e-acute
+    -- (\233, invalid as UTF-8) in the description, and a lone continuation
+    -- byte in the version
+    name = "\239\187\191Run Mode",
+    version = "1.0\128.0",
+    description = "caf\233 latt\233",
+    category = "UI\255" })
+  eq(m.name, "Run Mode", "a leading BOM is stripped from the name")
+  eq(m.version, "1.0.0", "invalid bytes are dropped from the version")
+  eq(m.description, "caf latt", "Latin-1 bytes are dropped, not replaced")
+  eq(m.raw.category, "UI", "raw.category is scrubbed in place for the badge")
+
+  local ok2 = mf({ id = "utf2", name = "Vers\195\163oVermelha", version = "1.0.0",
+    entry = "m.lua", description = "Pok\195\169mon \240\159\148\165" })
+  eq(ok2.name, "Vers\195\163oVermelha", "valid two-byte sequences survive")
+  eq(ok2.description, "Pok\195\169mon \240\159\148\165",
+    "valid three- and four-byte sequences survive")
+
+  -- surrogate half (ED A0 80) and overlong slash (C0 AF) are invalid even
+  -- though their lead bytes look plausible
+  local bad = mf({ id = "utf3", name = "a\237\160\128b\192\175c",
+    version = "1.0.0", entry = "m.lua" })
+  eq(bad.name, "abc", "surrogates and overlongs are dropped")
+end
+
 T.finish("launcher_mods")