Incorrect hiding of the string metatable

At the moment, in bios.lua, the string metatable is hidden like this:

local nativegetmetatable = getmetatable
local nativetype = type
local nativeerror = error
function getmetatable( _t )
	if nativetype( _t ) == "string" then
		nativeerror( "Attempt to access string metatable", 2 )
		return nil
	end
	return nativegetmetatable( _t )
end

This is a bad way to approach the problem - in normal lua, the contract of `getmetatable` is to never error. For instance, I might want to check if a value has a metatable, without wanting to do anything to it.

Lua provides a builtin way to protect metatables - the `__metatable` metafield:

getmetatable("").__metatable = "The string metatable is protected"

print(getmetatable("")) -- The string metatable is protected

They know about the __metatable field in general, beats me why the __metatable field isn't used for protecting the string metatable.

Thinking some more, it would be more useful to actually expose the string metatable in a readonly manner, by:

• Making a copy of the metatable. This has the advantage of being iterable, but the disadvantage of silently failing when the metatable is modified.

local stringmt = getmetatable("")
local clonedmt = {}
stringmt.__metatable = clonedmt
for k, v in pairs(stringmt) do
    clonedmt[k] = v
end

• Proxying access, firing an error on modification. Yes, rawset can be used, but that won't actually modify the real metatable. This has the disadvantage of making the table non-iterable, but the advantage of failing loudly.

local stringmt = getmetatable("")
stringmt.__metatable = setmetatable({}, {
    -- the `__index = stringmt` shorthand is unsuitable here
    __index = function(_, k)
        return stringmt[k]
    end,
    __newindex = function(_, k, v)
        nativeerror("The string metatable cannot be changed")
    end
})

The string metatable is shared between computers. I can't see your code on my phone, so I don't know if that affects it.

immibis, on 31 December 2012 - 02:25 AM said:

The string metatable is shared between computers.

Wait, really? Is that the only thing shared between computers/

Eric, on 02 January 2013 - 03:52 AM said:

immibis, on 31 December 2012 - 02:25 AM said:

The string metatable is shared between computers.

Wait, really? Is that the only thing shared between computers/

I think so. It's a static field in LuaJ so there's no easy way for CC to unshare it.

I don't see a reason to prevent getting the string metatable, it's not like the value can be changed outside of java

Except it can.

natedogith1, on 07 January 2013 - 09:48 PM said:

I don't see a reason to prevent getting the string metatable, it's not like the value can be changed outside of java

getmetatable("").__index.sub = os.reboot

That would reboot something whenever someone does aStringVariable:sub(…)
Not sure if it'll reboot the computer that put it in the metatable or the one running the sub command.

Cloudy, on 08 January 2013 - 12:25 AM said:

Except it can.

Not if you get it in a readonly way, like the two ways I suggested.

immibis, on 08 January 2013 - 12:48 AM said:

getmetatable("").__index.sub = os.reboot

That would reboot something whenever someone does aStringVariable:sub(…)
Not sure if it'll reboot the computer that put it in the metatable or the one running the sub command.

That particular example is no worse than `string.sub = os.reboot`, which is currently allowed - worse yet, all the bios code uses `string.func(s, …)` not `s:func(…)`, so this breaks things at a deeper level.

The case which would really break things is `getmetatable("").__concat = os.reboot`


Again, none of these are problems if the metatable is readonly (instead of inaccesible)

Why is that example no worse? One propagates to all computers - the other stays on one computer.

Cloudy, on 08 January 2013 - 08:17 AM said:

Why is that example no worse? One propagates to all computers - the other stays on one computer.

Whoops, forgot about the "The string metatable is shared between computers." bit

Ok, so I was being stupid and assumed that sense there was no setmetatable method, the metatable couldn't be changed

Eric, on 08 January 2013 - 07:45 AM said:

The case which would really break things is `getmetatable("").__concat = os.reboot`

IIRC all string operations are built-in except for indexing, and don't go through the metatable.