Re: wish for 5.2

[Mark Hamburg <mark@...>]

On Feb 13, 2009, at 3:44 AM, Roberto Ierusalimschy wrote:

> > My wish for userdata environments is for them to allow nil as the
> > environment. That way I could easily tell whether or not I had set a
> > custom environment on a userdata.
>
> But it would require an extra test for nil before any indexing. Can't
> you use a fixed table with (or without) a distinguishing field that
> you can inspect when you need to know whether there is a custom
> environment?

Saving a test when indexing is probably a win over doing the test when  
writing to see whether we need to generate a table. That's a good  
point that is a little too easy to forget when writing the code. It's  
been a while since I wrote any code dealing with this, so I can't  
recall whether there was some particular reason why sharing an empty  
table until a write occurred was problematic. The one other benefit I  
had thought of at the time was that it would save the GC from having  
to repeatedly trace the link to the shared empty table.

Mark