Re: [PROPOSAL] multiple keys/values with function metamethod
[
Date Prev][
Date Next][
Thread Prev][
Thread Next]
[
Date Index]
[
Thread Index]
- Subject: Re: [PROPOSAL] multiple keys/values with function metamethod
- From: Sean Conner <sean@...>
- Date: 2015年9月13日 00:01:39 -0400
It was thus said that the Great Rodrigo Azevedo once stated:
> I lost something here? The next phrase after my quote (not quoted here)
> says that this syntax is not good! (for pure tables)
Okay, so a table without a metatable will raise an error if you do
t1[a,b,c] = 1,2,3
but a table *with* a metatable *and* a __newindex() method will not. I
suppose that means that a table *with* a metatable that does *not* have
__newindex() will still error out. Okay, quick, which line will error out?
foo[a,b,c] = 1,2,3
bar[a,b,c] = 1,2,3
To be safe, you would have to do:
function getmetafield(t,field)
local mt = getmetatable(t)
if mt then
return mt[field]
end
end
if getmetafield(foo,"__newindex") then foo[a,b,c] = 1,2,3 end
if getmetafield(bar,"__newindex") then bar[a,b,c] = 1,2,3 end
Or you just do the former, let it error out, then swear as you track down
where bar was defined so you can add the appropriate metatable.
> I would like to stress that the proposal is not about multiple assigments
> or multidimensinal arrays. It is about to easily ( and fast, due to the
> extended use of the stack) implemention of these alternatives, as well as
> many more general cases.
But the problem here is that in one case,
foo[a,b,c] = 1,2,3
will generate an error, and in another case,
foo[a,b,c] = 1,2,3
will not. One of the nice qualities of Lua (and one of the more annoying
qualities of Lua, depending on my mood) is the lack of typechecking. Sure,
when I do this:
sock = net.socket(address.family,'tcp')
sock.reuseaddr = true
sock:bind(address)
sock is of type 'userdata'. But I can easily "fake" a socket:
sock = {}
sock.reuseaddr = false
function sock:bind(address) ... end
and now it's a table. The rest of my code doesn't care *how* it's defined,
as long as it *acts* like I expect it to act. Sure, I can make sure my
mocked sock has a __newindex() method, but without your proposal, I don't
have to bother with defining that.
Thus, if this:
foo[a,b,c] = 1,2,3
fails for non-metatable tables, then I would reject the proposal because it
breaks expected behavior.
-spc