Re: hpairs?

[Richard Hundt <richardhundt@...>]

On 12/29/2010 01:18 PM, Roberto Ierusalimschy wrote:
> > I use this for a JSON serializer to encode the hash part, and it
> > seems to work, but that may be implementation specific and not
> > reliable (i.e. it relies on next() returning the first key in the
> > hash part when given the last key in the array part). Critique?
>
> Just yesterday I sent an example where this does not work:
>
>
>    t = {a=1,b=2,c=3,d=5,e=6}; t[1] = 1; t[2] = 1; t[3] = 1
>
>    for k,v in next, t, #t>  0 and #t or nil do print(k) end
>
> Prints only '1' in my machine.

Okay, thanks, that's clearly not the way to do it then.

> The "array part" and the "hash part" do not correspond necessarily with
> the "sequence keys" (those returned by ipairs) and the "non-sequence
> keys".

Looking at the Table struct as well as luaH_next is obviously not the 
whole picture, so I'll take it as a fun exercise figuring out why that 
should be :)

> Please, do not refer to the "array part" or the "hash part" of a table,
> unless you are discussing the implementation of tables. They are
> completely invisible (except for performance).

Sorry, I'm probably an unusual case here... I spend most of my Lua 
coding time writing compilers which produce PUC-Rio Lua VM bytecode 
(relevant in this case is parametrising OP_NEWTABLE), so I'm biased to 
thinking in terms of the implementation, which can get tricky when 
implementation details leak through to semantics and you think that 
that's what's intended (which, in a way, is what my question was about).

I take your Point though.

Thanks,
Richard