Fixed injects into child classes

[Luke100000]

#2573

This change handles "cannot be injected" correctly if the field member is inherited.

---@class (exact) A
---@field test number
local a = {}

function a:init()
    --This is fine
    self.test = 0
end

---@class (exact) B : A
local b = {}

function b:init()
    --This is an injection and thus invalid
    self.test = 0

    --This is an actual injection and will remain invalid
    self.test2 = 0
end

[tomlau10]

The forloop here seems can early break 🤔

1. local class = vm.getDefinedClass(uri, source) doesn't depend on src, so it is a constant in this loop
2. Now whether class is nil or not, it is unrelated to outer loop, and checking the whole logic once is enough:
   • if the class has exact attribute => early return
   • if the class is nil or has no exact attribute => early break

        for _, src in ipairs(fields) do
            if src.type == "setfield" then
                local class = vm.getDefinedClass(uri, source)
                if class then
                    for _, doc in ipairs(class:getSets(uri)) do
                        if vm.docHasAttr(doc, 'exact') then
                            return
                        end
                    end
                end
                break  <-- the break here, because the above logic only need to check once
            end
        end

[Luke100000]

Ah yes, that's true, thanks. I will revisit this code (it's been a few months since I wrote that) and make a follow-up PR for that later.

[sumneko]

Thank you!