fix: LuaJIT ?? strips optional ? from the left-hand type

integer? stores nil as a node flag, so ?? never saw a nil object and
copied the optional type unchanged. Treat isOptional() as hasNil so
the existing removeOptional() branch runs.

Author: cpruijsen

changelog.md

@@ -2,6 +2,7 @@
 
 ## Unreleased
 <!-- Add all new changes here. They will be moved under a version at release -->
+* `FIX` LuaJIT nil-coalescing `??` now strips the optional `?` from the left-hand type (e.g. `integer? ?? 2` infers `integer`) [#3459](https://github.com/LuaLS/lua-language-server/issues/3459)
 * `FIX` A `.gitignore` line containing only whitespace no longer ignores the entire workspace [#3458](https://github.com/LuaLS/lua-language-server/issues/3458)
 * `FIX` `need-check-nil` diagnostic is no longer reported on safe navigation access (e.g. `x?.field`, `f?.()`, `t?.[key]`), since the optional access itself already handles the nil check. Note that a non-safe access chained after a safe one (e.g. `x.upper()?.field`) still reports, because the safe access only protects its own result.
 

script/vm/operator.lua

@@ -272,7 +272,7 @@ vm.binarySwitch = util.switch()
         local node2 = vm.compileNode(source[2])
         -- 统计具体类型：variable/local 是引用元信息，无具体类型时视为未知（可能为 nil）
         local count   = 0
-        local hasNil  = false
+        local hasNil  = node1:isOptional()
         for c in node1:eachObject() do
             if c.type == 'nil'
             or (c.type == 'global' and c.cate == 'type' and c.name == 'nil') then

test/type_inference/common.lua

@@ -3312,6 +3312,14 @@ end
 local <?x?> = n or 0
 ]]
 
+config.set(nil, 'Lua.runtime.nonstandardSymbol', { '??' })
+TEST 'integer' [[
+---@type integer?
+local foo = 1
+local <?bar?> = foo ?? 2
+]]
+config.set(nil, 'Lua.runtime.nonstandardSymbol', {})
+
 TEST 'number' [=[
 local <?x?> = F()--[[@as number]]
 ]=]